Cobolt Lasers - HÜBNER Photonics
bolt Lasers - HÜBNER Photonics Skip to content Search for: HomeProductsLaser Technology Diode lasersCobolt 06-01 SeriesSingle Frequency LasersCobolt 04-01 SeriesCobolt 05-01 SeriesCobolt 08-01 SeriesDPSS LasersCobolt 04-01 SeriesCobolt 05-01 SeriesCobolt 06-01 SeriesCobolt 08-01 SeriesCobolt Rogue SeriesNarrow linewidth lasersCobolt 08-01 SeriesNanosecond lasersCobolt Tor SeriesMulti-line lasersCobolt SkyraFiber LasersAmpheia™ Fiber Amplifier COMING SOONCustomized SolutionsCustomized lasersCustomized optical subsystemsFemtosecond lasersVALO SeriesTunable LasersCobolt Qu-T™ Series COMING SOONC-WAVE SeriesCobolt Odin SeriesLaser CombinersC-FLEXOptogenetics SolutionsOptions & AccessoriesDownloads (incl software)Terahertz Technology Terahertz spectrometersT-COGNITION®T-SPECTRALYZER® T/R/FTerahertz imagersT-SENSE®T-SENSE FMI Get help to choose! Product Selector Wavelength Guide320 nm355 nm375 nm395 nm405 nm415 nm425 nm445 nm450 nm457 nm473 nm488 nm491 nm505 nm515 nm520 nm532 nm553 nm561 nm594 nm633 nm638 nm640 nm647 nm650 nm660 nm685 nm705 nm707 nm730 nm760 nm780 nm785 nm808 nm813 nm830 nm915 nm940 nm975 nm1064 nm3264 nm3431 nm4330 nmTunable NIRTunable VISFemtosecond 1um Laser Technology – Wavelength Guide Read more Applications Laser Technology Biomedical ResearchDNA sequencingFluorescence MicroscopyMultiphoton microscopyFlow CytometryMALDI-TOFMicrodissectionOptogeneticsPhotoacoustics / OptoacousticsIndustrial MetrologyDynamic Light ScatteringHolographyInterferometryGas sensingLaser Doppler VelocimetryLIBSLIDARLithographyOptical TweezersPhotoacoustic spectroscopyRaman SpectroscopySemiconductor inspectionResearchLasers for Cold Atoms and MoleculesLithographyNanophotonicsQuantum TechnologiesSingle molecule spectroscopyMaterial ProcessingMarking and micromachining Terahertz Technology Public securityImaging of hidden objects and indentification of dangerous materialsQuality controlTHz for non-destructive testingFAQFAQKnowledge BankContactLocationsSales requestBook a meetingService & supportHamburger MenuAbout usCompany profilePrivacy policyEnvironmental initiativesHTCure manufacturing technologyCareersPress & EventsNewsTrade ShowsWebinarsVideosContactLocationsSales requestBook a meetingService & supportSocial mediaYoutubeLinkedinVimeoDownloads & OptionsDownloads (incl software)Options & Accessories Site map HomeProductsLaserDiode lasersCobolt 06-01 SeriesSingle Frequency LasersCobolt 04-01 SeriesCobolt 05-01 SeriesCobolt 08-01 SeriesDPSS LasersCobolt 04-01 SeriesCobolt 05-01 SeriesCobolt 06-01 SeriesCobolt 08-01 SeriesNarrow linewidth lasersCobolt 08-01 SeriesNanosecond lasersCobolt TorMulti-line lasersCobolt SkyraFiber LasersAzurlight SystemsCustomized SolutionsCustomized lasersCustomized optical subsystemsFemtosecond lasersVALO SeriesTunable LasersC-WAVE SeriesCobolt OdinLaser CombinersC-FLEXOptogenetics SolutionsOptions & AccessoriesDownloads (incl software)Terahertz TechnologyTerahertz spectrometersT-COGNITION®T-SPECTRALYZER® T/R/FTerahertz imagersT-SENSE®T-SENSE FMIWavelength Guide355 nm405 nm415 nm425 nm445 nm450 nm457 nm473 nm488 nm491 nm505 nm515 nm532 nm553 nm561 nm594 nm633 nm638 nm640 nm647 nm650 nm660 nm685 nm730 nm760 nm785 nm808 nm830 nm940 nm975 nm1064 nm3264 nm3431 nm4330 nmTunable NIRTunable VISFemtosecond 1umApplicationsLaser TechnologyBiomedical ResearchDNA sequencingFluorescence MicroscopyFlow CytometryMALDI-TOFMicrodissectionOptogeneticsPhotoacoustics / OptoacousticsIndustrial MetrologyDynamic Light ScatteringHolographyInterferometryLaser Doppler VelocimetryLIBSLIDARLithographyOptical TweezersPhotoacoustic spectroscopyRaman SpectroscopySemiconductor inspectionResearchGas sensingLasers for Cold Atoms and MoleculesLithographyNanophotonicsQuantum TechnologiesSingle molecule spectroscopyMaterial ProcessingMarking and micromachiningTerahertz TechnologyPublic securityImaging of hidden objects and indentification of dangerous materialsQuality controlTHz for non-destructive testingQ&AKnowledge BankAbout usCompany profilePrivacy policyEnvironmental initiativesHTCure manufacturing technologyVideosCareersPress & EventsNewsTrade showsContactHeadquarters and Sales OfficesSales and Information RequestsSign up for our newsletterEmail usSocial mediaLinkedinYoutubeVimeoDownloads & OptionsDownloads (incl software)Options & AccessoriesSitemap Search for: Home-Cobolt Lasers Cobolt Lasers2022-12-20T16:38:06+01:00 Explore our Cobolt lasersCobolt is now a part of HÜBNER Photonics Through the well-known Swedish laser manufacturer Cobolt AB, a proven supplier of high performance lasers of more than 20 years, HÜBNER Photonics division offers one of the industry’s broadest ranges of compact single-frequency CW lasers, diode lasers, multi-line lasers and Q-switched lasers across the full UV-Visible-NIR spectrum. Cobolt 04-01 SeriesSingle frequency, CW diode pumped lasers Wavelength: 457 nm – 1064 nm Power: 25 mW – 400 mW Applications: Raman, microscopy, LDV, DLS Read more Cobolt 05-01 SeriesHigh power, single frequency, CW diode pumped lasers Wavelength: 320 nm – 1064 nm Power: 10 mW – 3000 mW Applications: Holography, Raman, microscopy, flow cytometry, research Read more Cobolt 06-01 SeriesPlug & play modulated CW lasers Wavelength: 375 nm – 1064 nm Power: 40 mW – 400 mW Applications: Microscopy, flow cytometry, optogenetics Read more Cobolt Tor™ SeriesHigh performance Q-switched lasers Wavelength: 355, 532 and 1064 nm Pulse energy: 50 – 500 uJ/pulse (Single pulse – 7 kHz) Applications: LIBS, Maldi-TOF, Micro-machining & Marking Read more Cobolt Odin™ SeriesTunable Lasers Mid-IR Wavelength: 3264 nm – 4330 nm Power, pulse rate: 60 – 80 mW, 10 kHz Applications: PAS, LIBS, photoacoustics Read more Cobolt Skyra™A revolutionary multi-line laser platform Wavelength: 405 nm – 685 nm Power: 50 mW, 100 mW Applications: Microscopy, flow cytometry Read more Cobolt Rogue™ SeriesHigh Power, CW diode pumped lasers Wavelength: 640 nm Power: 1 W Applications: Super resolution microscopy Read more Get help to choose! If you are unsure of which Product, Series, Power or Wavelength you need, use our Product Selector. Product Selector Applications We expand the capabilities of your applications. Applications Sign up for our newsletter Do you want the latest news from us straight into your email? Sign up for our newsletter here. Register here Manufacturing sites Cobolt AB Visit: Vretenvägen 13 Delivery: Hemvärnsgatan 20 171 54 Solna, Sweden Phone: +46 8 545 912 30 Fax: +46 8 545 912 31 HÜBNER Photonics GmbH Wilhelmine-Reichard Strasse 6 34123 Kassel, Germany Phone: +49 561 994 060 – 0 Technical Support: +49 561 994 060 – 12 Fax: +49 561 994 060 – 13 Contact us General enquiries: info@hubner-photonics.com Sales enquiries: sales@hubner-photonics.com Here you can find contact information to all our offices: Contact page Also contact details for our distributors Sign up for our newsletter Follow us on LinkedIn Products Laser Technology Terahertz Technology Check out our Product Selector Organisation HÜBNER Photonics – a division of HÜBNER GmbH & Co. KG In the HÜBNER Photonics division, the principles of high-quality workmanship, superior reliability and lifetime are applied to the building of advanced tools and solutions. Find out more at: www.hubner-group.com Copyright © 2024 HÜBNER Photonics | All Rights Reserved | Privacy Policy | Cookies Page load link This website uses cookies to function optimally, and may process information about you in order to offer you marketing based on your use of the website, ie personalised advertising. Necessary Advertising Analytics Functionality Privacy Preferences I Agree Privacy Preference Center Options Consent Management Cookie Settings NecessaryAdvertisingAnalyticsFunctionality Privacy Policy Consent Management Cobolt collects, uses, discloses and stores some personal information. Cobolt is responsible for personal data and is responsible for processing personal data via this site and other contact with us. Our policy is updated from time to time. You can find the latest information by following the link below. Privacy Policy Required By using this site you agree to our privacy policy. Read the full policy here . Necessary Functional cookies are necessary for the website to function optimally. The purpose of these cookies is to track choices made by the visitor, eg. registering, using the webbshop or choosing a language. Read more about our use of cookies here Cookies Used Required gdpr[allowed_cookies], gdpr[consent_types] Advertising Advertising cookies collect statistics of how users interact with the website in order to be able to provide relevant advertisements based upon user experience. Read more about our use of cookies here Cookies Used ON OFF IDE Analytics Analytics cookies collect anonymous statistics on how visitors use the site. For example how many times the site is visited or how long the visitor stays. Cookies from Google Analytics are used, among other things, to differentiate users and collect statistics. Read more about our use of cookies here Cookies Used ON OFF cusid Other Functionality Functional cookies are necessary for the website to function optimally. The purpose of these cookies is to track choices made by the visitor, eg. registering, using the webbshop or choosing a language. Read more about our use of cookies here Cookies Used ON OFF vuid, GPS, PREF, VISITOR_INFO1_LIVE, YSC, remote_sid OK Book a meeting ✕ Requests Requesting a quotation or more product information has never been easier! Just send in your inquiry or request using our convenient Sales request form and we will make sure that a local sales representative gets back to you shortly. We use the personal data you share with us to solely respond to your sales enquiry either through our direct offices or via our distributor network. See our privacy policy for details on how we handle your personal data. ✕ Email us ✕ Cobolt is now a part of HÜBNER PhotonicsThrough the well-known Swedish laser manufacturer Cobolt AB, a proven supplier of high performance lasers of more than 15 years, HÜBNER Photonics division offers one of the industry’s broadest ranges of compact single-frequency CW lasers, diode lasers, multi-line lasers and Q-switched lasers across the full UV-Visible-NIR spectrum. Continue to HUBNER Photonics ✕ Requests Requesting a quotation or more product information has never been easier! Just send in your inquiry or request using our convenient Sales request form and we will make sure that a local sales representative gets back to you shortly. We use the personal data you share with us to solely respond to your sales enquiry either through our direct offices or via our distributor network. See our privacy policy for details on how we handle your personal data. ✕ Requests Requesting a quotation or more product information has never been easier! Just send in your inquiry or request using our convenient Sales request form and we will make sure that a local sales representative gets back to you shortly. We use the personal data you share with us to solely respond to your sales enquiry either through our direct offices or via our distributor network. See our privacy policy for details on how we handle your personal data. ✕ Requests Requesting a quotation or more product information has never been easier! Just send in your inquiry or request using our convenient Sales request form and we will make sure that a local sales representative gets back to you shortly. We use the personal data you share with us to solely respond to your sales enquiry either through our direct offices or via our distributor network. See our privacy policy for details on how we handle your personal data. ✕ Email us ✕ Requests Requesting a quotation or more product information has never been easier! Just send in your inquiry or request using our convenient Sales request form and we will make sure that a local sales representative gets back to you shortly. We use the personal data you share with us to solely respond to your sales enquiry either through our direct offices or via our distributor network. See our privacy policy for details on how we handle your personal data. ✕ Requests Requesting a quotation or more product information has never been easier! Just send in your inquiry or request using our convenient Sales request form and we will make sure that a local sales representative gets back to you shortly. We use the personal data you share with us to solely respond to your sales enquiry either through our direct offices or via our distributor network. See our privacy policy for details on how we handle your personal data. ✕ Request a quote Go to TopCobolt激光器 - 北京鼎信优威光子科技有限公司
Cobolt激光器 - 北京鼎信优威光子科技有限公司
跳过内容
首页资讯中心产品中心资料下载联系我们
搜索：
载入中...
来自瑞典高性能DPSS激光器
HTCure™技术打造DPSS激光器近乎完美光斑品质和光束质量
低噪声，高稳定性超长使用寿命，最长支持2年质保
Cobolt激光器dynadmin2020-05-12T14:45:30+08:00
产品列表dynadmin2023-09-13T10:30:25+08:00
VALO 飞秒光纤激光器
VALO 飞秒光纤激光器dynadmin2020-05-09T03:15:08+08:00
C-FLEX激光合束器
C-FLEX激光合束器dynadmin2021-06-14T22:50:14+08:00
COBOLT 调Q纳秒激光器
COBOLT 调Q纳秒激光器dynadmin2021-06-14T22:49:11+08:00
COBOLT 中红外OPO调谐激光器
COBOLT 中红外OPO调谐激光器dynadmin2020-05-09T03:15:11+08:00
C-WAVE OPO调谐激光器
C-WAVE OPO调谐激光器dynadmin2023-09-13T10:32:18+08:00
COBOLT激光器Skyra系列
COBOLT激光器Skyra系列dynadmin2023-09-13T10:34:09+08:00
COBOLT激光器06-01系列
COBOLT激光器06-01系列dynadmin2023-09-13T10:33:33+08:00
COBOLT激光器08-01系列
COBOLT激光器08-01系列dynadmin2020-05-09T03:15:14+08:00
COBOLT激光器05-01系列
COBOLT激光器05-01系列波长选择指南355nm405nm445nm473nm488nm491nm515nm532nm553nm561nm594nm633nm638nm640nm647nm660nm785nm1064nmMid-IR
© Copyright 2004 - | 北京鼎信优威光子科技有限公司 京ICP备05074267号-3
Page load link
返回顶部
COBOLT激光器04-01系列 - 北京鼎信优威光子科技有限公司
COBOLT激光器04-01系列 - 北京鼎信优威光子科技有限公司
跳过内容
首页资讯中心产品中心资料下载联系我们
搜索：
COBOLT激光器04-01系列dynadmin2020-05-09T03:15:14+08:00
Cobolt 04-01 Series
04-01系列紧凑型、低噪声、单纵模DPSS连续激光器
• 457-1064 nm, 连续输出功率高达 400 mW
• 理想的 TEM00 光束
• 极其坚固紧实的密封封装
• 低噪声, <0.25% rms
• 可选尾纤输出（激光头内部光纤耦合）
• 可选集成式调制模块，频率高达 3MHz
产品特性产品参数DataSheet视频资料产品特性
高温固化处理(HTCure)的超坚固耐用DPSS激光器
瑞典Cobolt公司采用HTCure™技术打造DPSS激光器，近乎理想光斑品质和光束质量，超强的稳定性和超长使用寿命。同时一改激光器作为敏感光学仪器的形象，以其皮实坚固的特点成为众多激光分析仪器的首选。
当今的高科技诊断仪器及科学实验对于激光器的性能期望值越来越高，作为代表性的紧凑型DPSS激光器越来越广泛的应用于这些系统中，以代替传统的氩离子激光器和灯泵激光。如DNA测序仪器经常性的单次运行好几天，以得到一个基因序列的据，此时激光器的任何一个问题都可能造成测试数据无效。基于这种情况，激光器的制造厂商需要制造出数年连续使用无差错的产品以满足需要，实际上这非常严苛。激光器是一个非常柔弱和敏感的仪器，即使是其中的光学器件表面被污染，抑或是谐振腔出现μm或亚mrad的机械位移，都会对激光器的性能大打折扣。
DPSS vs 气体激光器
DPSS激光器的固体工作物质，可以提供更长的寿命。但即使如此，如果激光器里的组件不够稳定、热敏感性及震动敏感性强，其相对与气体激光器的优势就会被抹杀。
Cobolt独有的HTCure技术
材料的选定：
为保持良好的温度稳定性，每一个组件的设计非常小心，其中最重要的就是匹配元器件单独、及相互之间的热膨胀系数。基于精心的器件选配，整个激光头可以被加热至150℃且保持稳定的相对位置，为高温固化技术（HTCure）提供了基础。
HTCure技术：
相对于传统的紫外光固化胶合方式，Cobolt采用了的高温固化方式来固定谐振腔，铸就了更可靠的封装，避免了传统工艺中的漏气问题，以及长期使用中的位置飘移问题，让激光器的稳定性大大提升。激光头的制作从泵浦激光及相应的准直、聚焦器件的固定开始，接着是激光晶体的准直、产生TEM00模的腔镜准直、PPKTP晶体（频率转换）的准直，最后是激光腔内外的其他各种元器件的准直。各部分之间的位置先暂时用紫外固化胶固定，然后整个激光头放在一个特制的烤箱中加热，完成一个完全的高温固化过程。
封装：
为了保证脆弱的光学元器件不受污染，固化后的激光头需要作密封封装，操作是在氮气环境下进行的，避免了将湿气封装的危险，保障了激光器的长久可用性和室内外不同环境的可用性。
HTCure的优势：
采用HTCure技术制造的激光器，比传统工艺的产品具备更高的准直精度和更严格的固定，强度的稳定性更好，噪声更低，尤其重要的是外部环境的扰动对其性能无法造成任何负面影响（激光器可暴露在超过100℃的环境下）。
温度扰动测试
图1：上左图：HTCure工艺的激光器在20-50℃温度扰动下的噪声表现； 上右图： HTCure工艺的激光器在20-50℃温度扰动，及快速开关下的功率稳定性表现。下图：HTCure工艺的激光器在20-50℃温度扰动下，指向稳定性的测试数据（< 4 μrad/℃）。
震动-抗冲击测试
图2：全系列产品测试按照IEC 600 68-X标准；可承受60G冲击（8ms）；可承受1m的自由落体冲击。
寿命测试：
图3：HTCure工艺的激光器寿命测试数据
Cobolt激光器，已作为可靠的光源被广泛应用于各个领域中，尤其是生命科学领域和激光拉曼系统中。
产品参数
波长和功率
产品名称
波长
功率
尾纤选项
调制选项
Cobolt Twist™
457 nm
≤ 50 mW
–
–
Cobolt Blues™
473 nm
≤ 50 mW
≤ 35 mW
≤ 40 mW
Cobolt Calypso™
491 nm
≤ 100 mW
≤ 100 mW
≤ 80 mW
Cobolt Fandango™
515 nm
≤ 150 mW
≤ 100 mW
≤ 120 mW
Cobolt Samba™
532 nm
≤ 300 mW
≤ 150 mW
≤ 240 mW
Cobolt Jive™
561 nm
≤ 200 mW
≤ 100 mW
≤ 120 mW
Cobolt Mambo™
594 nm
≤ 100 mW
≤ 75 mW
≤ 80 mW
Cobolt Flamenco™
660 nm
≤ 100 mW
–
–
Cobolt Rumba™
1064 nm
≤ 400 mW
–
–
其他主要参数
光谱线宽
<1MHz
波长稳定性（预热后）
2 pm over ± 2 ºC and 8 hrs
光斑模式
TEM00 M2 <1.1
光束直径（出光口处）
700 μm ± 50 μm
光束圆度（出光口处）
>0.95 : 1
光束发散角（全角）
<1.2 mrad（<1.3mrad @ 594 nm ）
长期稳定性（8 hrs ±3°C）
<2%（<3% for Calypso, Mambo）
偏振比（linear, vertical）
>100:1
尺寸
激光头：102 x 60 x 40 mm
控制器：190 x 72 x 28 mm
DataSheet
视频资料
Sorry, your browser doesn't support embedded videos.
在线留言获取报价联系我们
波长选择指南
355nm405nm445nm473nm488nm491nm515nm532nm553nm561nm594nm633nm638nm640nm647nm660nm785nm1064nmMid-IR
© Copyright 2004 - | 北京鼎信优威光子科技有限公司 京ICP备05074267号-3
Page load link
返回顶部
Cobolt 06-01 Series - HÜBNER Photonics
Cobolt 06-01 Series - HÜBNER Photonics
Skip to content
Search for:
HomeProductsLaser Technology
Diode lasersCobolt 06-01 SeriesSingle Frequency LasersCobolt 04-01 SeriesCobolt 05-01 SeriesCobolt 08-01 SeriesDPSS LasersCobolt 04-01 SeriesCobolt 05-01 SeriesCobolt 06-01 SeriesCobolt 08-01 SeriesCobolt Rogue SeriesNarrow linewidth lasersCobolt 08-01 SeriesNanosecond lasersCobolt Tor SeriesMulti-line lasersCobolt SkyraFiber LasersAmpheia™ Fiber Amplifier COMING SOONCustomized SolutionsCustomized lasersCustomized optical subsystemsFemtosecond lasersVALO SeriesTunable LasersCobolt Qu-T™ Series COMING SOONC-WAVE SeriesCobolt Odin SeriesLaser CombinersC-FLEXOptogenetics SolutionsOptions & AccessoriesDownloads (incl software)Terahertz Technology
Terahertz spectrometersT-COGNITION®T-SPECTRALYZER® T/R/FTerahertz imagersT-SENSE®T-SENSE FMI
Get help to choose!
Product Selector
Wavelength Guide320 nm355 nm375 nm395 nm405 nm415 nm425 nm445 nm450 nm457 nm473 nm488 nm491 nm505 nm515 nm520 nm532 nm553 nm561 nm594 nm633 nm638 nm640 nm647 nm650 nm660 nm685 nm705 nm707 nm730 nm760 nm780 nm785 nm808 nm813 nm830 nm915 nm940 nm975 nm1064 nm3264 nm3431 nm4330 nmTunable NIRTunable VISFemtosecond 1um
Laser Technology – Wavelength Guide
Read more
Applications
Laser Technology
Biomedical ResearchDNA sequencingFluorescence MicroscopyMultiphoton microscopyFlow CytometryMALDI-TOFMicrodissectionOptogeneticsPhotoacoustics / OptoacousticsIndustrial MetrologyDynamic Light ScatteringHolographyInterferometryGas sensingLaser Doppler VelocimetryLIBSLIDARLithographyOptical TweezersPhotoacoustic spectroscopyRaman SpectroscopySemiconductor inspectionResearchLasers for Cold Atoms and MoleculesLithographyNanophotonicsQuantum TechnologiesSingle molecule spectroscopyMaterial ProcessingMarking and micromachining
Terahertz Technology
Public securityImaging of hidden objects and indentification of dangerous materialsQuality controlTHz for non-destructive testingFAQFAQKnowledge BankContactLocationsSales requestBook a meetingService & supportHamburger MenuAbout usCompany profilePrivacy policyEnvironmental initiativesHTCure manufacturing technologyCareersPress & EventsNewsTrade ShowsWebinarsVideosContactLocationsSales requestBook a meetingService & supportSocial mediaYoutubeLinkedinVimeoDownloads & OptionsDownloads (incl software)Options & Accessories
Site map
HomeProductsLaserDiode lasersCobolt 06-01 SeriesSingle Frequency LasersCobolt 04-01 SeriesCobolt 05-01 SeriesCobolt 08-01 SeriesDPSS LasersCobolt 04-01 SeriesCobolt 05-01 SeriesCobolt 06-01 SeriesCobolt 08-01 SeriesNarrow linewidth lasersCobolt 08-01 SeriesNanosecond lasersCobolt TorMulti-line lasersCobolt SkyraFiber LasersAzurlight SystemsCustomized SolutionsCustomized lasersCustomized optical subsystemsFemtosecond lasersVALO SeriesTunable LasersC-WAVE SeriesCobolt OdinLaser CombinersC-FLEXOptogenetics SolutionsOptions & AccessoriesDownloads (incl software)Terahertz TechnologyTerahertz spectrometersT-COGNITION®T-SPECTRALYZER® T/R/FTerahertz imagersT-SENSE®T-SENSE FMIWavelength Guide355 nm405 nm415 nm425 nm445 nm450 nm457 nm473 nm488 nm491 nm505 nm515 nm532 nm553 nm561 nm594 nm633 nm638 nm640 nm647 nm650 nm660 nm685 nm730 nm760 nm785 nm808 nm830 nm940 nm975 nm1064 nm3264 nm3431 nm4330 nmTunable NIRTunable VISFemtosecond 1umApplicationsLaser TechnologyBiomedical ResearchDNA sequencingFluorescence MicroscopyFlow CytometryMALDI-TOFMicrodissectionOptogeneticsPhotoacoustics / OptoacousticsIndustrial MetrologyDynamic Light ScatteringHolographyInterferometryLaser Doppler VelocimetryLIBSLIDARLithographyOptical TweezersPhotoacoustic spectroscopyRaman SpectroscopySemiconductor inspectionResearchGas sensingLasers for Cold Atoms and MoleculesLithographyNanophotonicsQuantum TechnologiesSingle molecule spectroscopyMaterial ProcessingMarking and micromachiningTerahertz TechnologyPublic securityImaging of hidden objects and indentification of dangerous materialsQuality controlTHz for non-destructive testingQ&AKnowledge BankAbout usCompany profilePrivacy policyEnvironmental initiativesHTCure manufacturing technologyVideosCareersPress & EventsNewsTrade showsContactHeadquarters and Sales OfficesSales and Information RequestsSign up for our newsletterEmail usSocial mediaLinkedinYoutubeVimeoDownloads & OptionsDownloads (incl software)Options & AccessoriesSitemap
Search for:
Home-Products-Cobolt 06-01 Series
Request a quote
Cobolt 06-01 Series
Plug & play modulated CW lasers
Cobolt 06-01 Series offers lasers over a large wavelength range in a compact plug and play format. The Series consists of high performance fixed wavelength diode laser modules (MLD) and diode-pumped lasers (DPL). High speed direct modulation capability and true off during modulation makes them ideal for applications in bioimaging and quantum technologies.
375 nm – 1064 nm, up to 400 mW
Fully integrated electronics
Integrated clean-up filter on all diode wavelengths
Digital & analog modulation with true OFF during modulation (>70 dB)
Fiber pigtailed configuration available
Compatible in the C-FLEX Laser Combiner
Ultra-robust, proven reliability
24 months warranty and < 2 weeks standard lead time
Applications
Perfect for fluorescence microscopy, flow cytometry, optogenetics, colorimetry, quantum technology, DNA sequencing.
Categories: 1064 nm laser, 375 nm laser, 395 nm laser, 405 nm laser, 415 nm laser, 425 nm laser, 445 nm laser, 457 nm laser, 473 nm laser, 488 nm laser, 505 nm laser, 515 nm laser, 520 nm laser, 532 nm laser, 553 nm laser, 561 nm laser, 633 nm laser, 638 nm laser, 647 nm laser, 660 nm laser, 685 nm laser, 690 nm laser, 705 nm laser, 730 nm laser, 760 nm laser, 785 nm laser, 808 nm laser, 830 nm laser, 915 nm laser, 940 nm laser, 975 nm laser, Diode Lasers, Lasers
Tags: 1064 nm, 375, 395 nm, 405 nm, 405nm, 405nm diode lasers, 445 nm, 445nm diode lasers, 473 nm, 473nm, 473nm diode lasers, 488 nm, 488nm, 488nm diode laser, 515 nm, 515nm diode lasers, 532 nm, 532nm, 532nm DPL, 553 nm, 553nm, 553nm diode pumped laser, 553nm laser, 561 nm, 561nm, 561nm DPL, 633 nm, 633nm, 633nm diode lasers, 638 nm, 638nm, 647 nm, 647nm, 647nm diode lasers, 660 nm, 660nm, 660nm diode lasers, blue diode lasers, cobolt 06, cobolt 06 lasers, cobolt 06-01, diode lasers, DPL, lasers for confocal microscopy, lasers for flow cytometry, lasers for fluorescence microscopy, lasers for super resolution microscopy, MLD, modulated diode laser, modulated laser, plug and play, red diode lasers
Specifications
Performance Data
Data sheets
Manuals
Literature
Software
Drawings
Videos
Options & Accessories
Specifications*Prices include everything you need to get started for lab use (CDRH compliant): power supply, key box, cables and even an integrated clean-up filter! (ex shipping/handling)
Fiber coupling options are available for multimode and single mode fibers using FIC-05
Product
Wavelength
Power
Fiber pigtailed option
Modulation speed
(rise time)
Price from*
Cobolt 06-MLD
375 nm
70 mW
25 mW
150 MHz (<2.5 ns)
€ 6,650 $ 7,170
£ 5,790
Cobolt 06-MLD
395 nm
120 mW
25 mW
150 MHz (<2.5 ns)
€ 6,900 $ 7,440
£ 6,000
Cobolt 06-MLD
405 nm
150 mW
365 mW
75 mW
150 mW
150 MHz (<2.5 ns)
€ 2,750 $ 2,970
£ 2,390
Cobolt 06-MLD
415 nm
120 mW
60 mW
150 MHz (<2.5 ns)
€ 7,400 $ 7,980
£ 6,440
Cobolt 06-MLD
425 nm
120 mW
60 mW
150 MHz (<2.5 ns)
€ 7,400 $ 7,980
£ 6,440
Cobolt 06-MLD
445 nm
100 mW
400 mW
50 mW
150 mW
150 MHz (<2.5 ns)
€ 5,350 $ 5,770
£ 4,650
Cobolt 06-MLD
457 nm
100 mW
400 mW
50 mW
150 mW
150 MHz (<2.5 ns)
€ 5,200 $ 5,610
£ 4,520
Cobolt 06-MLD
473 nm
100 mW
300 mW
50 mW
150 mW
150 MHz (<2.5 ns)
€ 5,350 $ 5,770
£ 4,650
Cobolt 06-MLD
488 nm
60 mW
100 mW
150 mW
200 mW
300 mW NEW!
30 mW
100 mW
150 MHz (<2.5 ns)
€ 4,200 $ 4,530
£ 3,650
Cobolt 06-MLD
505 nm
80 mW
40 mW
150 MHz (<2.5 ns)
€ 6,650 $ 7,170
£ 5,790
Cobolt 06-MLD
515 nm
80 mW
150 mW
40 mW
75 mW
150 MHz (<2.5 ns)
€ 5,750 $ 6,200
£ 5,000
Cobolt 06-MLD
520 nm
80 mW
40 mW
150 MHz (<2.5 ns)
€ 3,200 $ 3,450
£ 2,780
Cobolt 06-DPL
532 nm
25 mW
50 mW
100 mW
200 mW
300 mW
400 mW
25 mW
50 mW
100 mW
200 mW
<50 kHz
(<6 us)
€ 5,100 $ 5,500
£ 4,440
Cobolt 06-DPL
553 nm
25 mW
50 mW
25 mW
<5 kHz
(<60 us)
€ 5,900 $ 6,360
£ 5,130
Cobolt 06-DPL
561 nm
25 mW
50 mW
100 mW
200 mW
25 mW
50 mW
100 mW
<10 kHz
(<30 us)
€ 5,900 $ 6,360
£ 5,130
Cobolt 06-DPL
594 nm NEW!
50 mW
100 mW
-
_
TBD
Cobolt 06-MLD
633 nm
80 mW
40 mW
150 MHz (<2.5 ns)
€ 2,600 $ 2,800
£ 2,260
Cobolt 06-MLD
638 nm
180 mW
80 mW
150 MHz (<2.5 ns)
€ 2,450 $ 2,640
£ 2,130
Cobolt 06-MLD
647 nm
130 mW
60 mW
150 MHz (<2.5 ns)
€ 3,700 $ 3,990
£ 3,220
Cobolt 06-MLD
660 nm
100 mW
50 mW
150 MHz (<2.5 ns)
€ 2,150 $ 2,320
£ 1,870
Cobolt 06-MLD
685 nm
40 mW
20 mW
150 MHz (<2.5 ns)
€ 2,500 $ 2,700
£ 2,180
Cobolt 06-MLD
690 nm
200 mW
75 mW
150 MHz (<2.5 ns)
€ 6,300 $ 6,790
£ 5,480
Cobolt 06-MLD
705 nm
30 mW
15 mW
150 MHz (<2.5 ns)
€ 3,000 $ 3,240
£ 2,610
Cobolt 06-MLD
730 nm
50 mW
20 mW
150 MHz (<2.5 ns)
€ 3,100 $ 3,340
£ 2,700
Cobolt 06-MLD
760 nm
25 mW
15 mW
150 MHz (<2.5 ns)
€ 4,600 $ 4,960
£ 4,000
Cobolt 06-MLD
785 nm
250 mW
100 mW
150 MHz (<2.5 ns)
€ 2,750 $ 2,970
£ 2,390
Cobolt 06-MLD
808 nm
120 mW
50 mW
150 MHz (<2.5 ns)
€ 2,800 $ 3,020
£ 2,440
Cobolt 06-MLD
830 nm
250 mW
100 mW
150 MHz (<2.5 ns)
€ 3,000 $ 3,240
£ 2,610
Cobolt 06-MLD
852 nm
50 mW
20 mW
150 MHz (<2.5 ns)
€ 2,530 $ 2,730
£ 2,200
Cobolt 06-MLD
915 nm
250 mW
100 mW
150 MHz (<2.5 ns)
€ 2,950 $ 3,180
£ 2,570
Cobolt 06-MLD
940 nm
250 mW
75 mW
150 MHz (<2.5 ns)
€ 2,990 $ 3,220
£ 2,600
Cobolt 06-MLD
975 nm
250 mW
50 mW
150 MHz (<2.5 ns)
€ 2,750 $ 2,970
£ 2,390
Cobolt 06-MLD
1064 nm
200 mW
75 mW
150 MHz (<2.5 ns)
€ 3,250 $ 3,500
£ 2,830
*Please note that Local currency may change. Shipping, tax etc. is not included in the displayed price. Displayed pricing denotes lowest power model.
Performance Data (click to enlarge)
Data Sheets
Datasheet Cobolt 06-01 SeriesPlug & Play Modulatable – Continuous Wave lasers
Manuals
Cobolt 06-01 Series:Plug & play modulated CW lasers
Literature (see also applications specific pages)
References
Application
Publication
Technology leaps in quantum sensing
Quantum sensing
White paper
B. Kemper Digital holographic microscopy PhotonicsViews 2020
Digital holographic microscopy
Photonics Views Magazine Jan 2020
TRAST microscopy for measuring oxygen concentration
Microscopy
Application note
Digital Holographic Microscopy for Label-Free Detection of Leukocyte Alternations Associated with Perioperative Inflammation after Cardiac Surgery
Digital holographic microscopy
Cells 2022
Label-Free Digital Holographic Microscopy for In Vitro Cytotoxic Effect Quantification of Organic Nanoparticles
Digital holographic microscopy
Cells 2022
Software
Software
Description
Cobolt Monitor
Cobolt MonitorTM is a combined GUI for all Cobolt laser models.The GUI supports simultaneous connection and monitoring of multiple lasers,configured for either RS232 or USB2.0. The GUI automatically searches fornew Cobolt lasers connected to the computer and presents them listed.
USB Driver
To connect to the laser via USB, the USB driver must be installed.When installed, a virtual COM port will be created to communicatewith the laser.
Lab View
Use to integrate the Cobolt lasers into Lab view environment.
GitHub
See a list of suggested code in C++ or Python on GitHub.
Integration into imaging softwarer
A Micro-manager driver is available for Cobolt 06-01 lasers.Please search for and install the “Cobolt Official” driver from theMicro-manager library.The driver is optimized for Cobolt 06-01 Series lasers manufacturedafter June 2020 but also back-compatible with lasers manufacturedbefore then.
If you experience any difficulties with your Micromanager integration,please contact us at: info@coboltlasers.com.
Mechanical specifications (click to enlarge)
Click here to register and download the Cobolt 06-01 CAD drawing (.STEP file)
Getting Started with Cobolt 06-01 Series
Cobolt 06-01 Series Laser – Unboxing – Long version
Options & Accessories
Cobolt FIC-05
HS-03 Laser Head Heatsink
Cobolt TEC-plate
Mounting plate for fiber coupling
Heatsink without fans
Temperature controlled mounting plate
See more options and accessories
If you are unsure of which Product, Series, Power or Wavelength you need, use our Product Selector.
Product Selector
Do you have a question?Our laser experts are here to help you!
Email us
Related productsC-FLEXThe compact and flexible laser combinerOptogenetics SolutionsSpecifically tailored for advanced Optogenetics researchCobolt Skyra™A revolutionary multi-line laser platform
Manufacturing sites Cobolt AB
Visit: Vretenvägen 13
Delivery: Hemvärnsgatan 20
171 54 Solna, Sweden
Phone: +46 8 545 912 30
Fax: +46 8 545 912 31
HÜBNER Photonics GmbH
Wilhelmine-Reichard Strasse 6
34123 Kassel, Germany
Phone: +49 561 994 060 – 0
Technical Support: +49 561 994 060 – 12
Fax: +49 561 994 060 – 13
Contact us General enquiries: info@hubner-photonics.com
Sales enquiries: sales@hubner-photonics.com
Here you can find contact information to all our offices: Contact page
Also contact details for our distributors
Sign up for our newsletter
Follow us on LinkedIn
Products Laser Technology
Terahertz Technology
Check out our Product Selector
Organisation HÜBNER Photonics
– a division of HÜBNER GmbH & Co. KG
In the HÜBNER Photonics division, the principles of high-quality workmanship, superior reliability and lifetime are applied to the building of advanced tools and solutions.
Find out more at:
www.hubner-group.com
Copyright © 2024 HÜBNER Photonics | All Rights Reserved | Privacy Policy | Cookies
Page load link
This website uses cookies to function optimally, and may process information about you in order to offer you marketing based on your use of the website, ie personalised advertising.
Necessary
Advertising
Analytics
Functionality
Privacy Preferences
I Agree
Privacy Preference Center
Options
Consent Management
Cookie Settings
NecessaryAdvertisingAnalyticsFunctionality
Privacy Policy
Consent Management
Cobolt collects, uses, discloses and stores some personal information. Cobolt is responsible for personal data and is responsible for processing personal data via this site and other contact with us. Our policy is updated from time to time. You can find the latest information by following the link below.
Privacy Policy
Required
By using this site you agree to our privacy policy. Read the full policy here .
Necessary
Functional cookies are necessary for the website to function optimally. The purpose of these cookies is to track choices made by the visitor, eg. registering, using the webbshop or choosing a language.
Read more about our use of cookies here
Cookies Used
Required
gdpr[allowed_cookies], gdpr[consent_types]
Advertising
Advertising cookies collect statistics of how users interact with the website in order to be able to provide relevant advertisements based upon user experience.
Read more about our use of cookies here
Cookies Used
ON
OFF
IDE
Analytics
Analytics cookies collect anonymous statistics on how visitors use the site. For example how many times the site is visited or how long the visitor stays.
Cookies from Google Analytics are used, among other things, to differentiate users and collect statistics.
Read more about our use of cookies here
Cookies Used
ON
OFF
cusid
Other
Functionality
Functional cookies are necessary for the website to function optimally. The purpose of these cookies is to track choices made by the visitor, eg. registering, using the webbshop or choosing a language.
Read more about our use of cookies here
Cookies Used
ON
OFF
vuid, GPS, PREF, VISITOR_INFO1_LIVE, YSC, remote_sid
OK
Cobolt 06-01 CAD drawings
✕
Book a meeting
✕
Requests
Requesting a quotation or more product information has never been easier! Just send in your inquiry or request using our convenient Sales request form and we will make sure that a local sales representative gets back to you shortly.
We use the personal data you share with us to solely respond to your sales enquiry either through our direct offices or via our distributor network. See our privacy policy for details on how we handle your personal data.
✕
Email us
✕
Cobolt is now a part of HÜBNER PhotonicsThrough the well-known Swedish laser manufacturer Cobolt AB, a proven supplier of high performance lasers of more than 15 years, HÜBNER Photonics division offers one of the industry’s broadest ranges of compact single-frequency CW lasers, diode lasers, multi-line lasers and Q-switched lasers across the full UV-Visible-NIR spectrum.
Continue to HUBNER Photonics
✕
Requests
Requesting a quotation or more product information has never been easier! Just send in your inquiry or request using our convenient Sales request form and we will make sure that a local sales representative gets back to you shortly.
We use the personal data you share with us to solely respond to your sales enquiry either through our direct offices or via our distributor network. See our privacy policy for details on how we handle your personal data.
✕
Requests
Requesting a quotation or more product information has never been easier! Just send in your inquiry or request using our convenient Sales request form and we will make sure that a local sales representative gets back to you shortly.
We use the personal data you share with us to solely respond to your sales enquiry either through our direct offices or via our distributor network. See our privacy policy for details on how we handle your personal data.
✕
Requests
Requesting a quotation or more product information has never been easier! Just send in your inquiry or request using our convenient Sales request form and we will make sure that a local sales representative gets back to you shortly.
We use the personal data you share with us to solely respond to your sales enquiry either through our direct offices or via our distributor network. See our privacy policy for details on how we handle your personal data.
✕
Email us
✕
Requests
Requesting a quotation or more product information has never been easier! Just send in your inquiry or request using our convenient Sales request form and we will make sure that a local sales representative gets back to you shortly.
We use the personal data you share with us to solely respond to your sales enquiry either through our direct offices or via our distributor network. See our privacy policy for details on how we handle your personal data.
✕
Requests
Requesting a quotation or more product information has never been easier! Just send in your inquiry or request using our convenient Sales request form and we will make sure that a local sales representative gets back to you shortly.
We use the personal data you share with us to solely respond to your sales enquiry either through our direct offices or via our distributor network. See our privacy policy for details on how we handle your personal data.
✕
Request a quote
Go to Top
Cobolt：如何为拉曼光谱选择激光器？ - 知乎
Cobolt：如何为拉曼光谱选择激光器？ - 知乎首发于科技翻译切换模式写文章登录/注册Cobolt：如何为拉曼光谱选择激光器？RealSharp这个人什么都不懂，但他什么都想问问。拉曼光谱有很多不同的波长，从紫外到可见光到近红外。对于某种给定的应用，如何选择最佳照明波长有时并不明显，因为拉曼系统的优化需要考虑很多因素，其中有些就与波长选择直接相关。首先，拉曼信号非常微弱，它取决于样品材料中光子间的相互作用，一般概率只有百万分之一。另外，拉曼散射强度和照明波长的四次方成反比，所以随着波长变长，拉曼信号迅速减弱。其次，探测灵敏度也和波长范围有关。无制冷硅基CCD器件的量子效率在800 nm后急剧下降。长波长可使用铟镓砷(InGaAs)阵列器件，如图1所示；不过噪声更大，灵敏度更低，成本也更高。空间分辨率也是考虑因素，因为成像分辨率受照明波长影响，衍射极限光斑约等于0.3λ。图1. Si和InGaAs探测器的灵敏度曲线短波长激光：VIS还是UV？由于拉曼信号强度、探测灵敏度和光谱分辨率都与波长有关，短波长(UV和VIS)看似比长波长(NIR)更适合。但是短波长照明需要克服一个关键问题：荧光。很多材料可能受UV-VIS光刺激而产生荧光，而荧光背景可能淹没更弱的拉曼信号。避免荧光影响的方法之一是移频激发拉曼差分光谱(SERDS)。这种方法记录两紧邻激光线的拉曼光谱，然后彼此相减。由于大部分荧光信号受波长影响不大，故能以此消除荧光影响。对于某些材料，使用更短的UV波长能够有效避免荧光影响，如图2所示。虽然UV光一般会激发强荧光，但是发射的荧光一般都在300 nm以上。所以，在300 nm以下激发和记录拉曼信号也是一种可行方法，因为拉曼信号非常接近照明波长。图2. 使用短波UV照明避免荧光影响但是，选择照明波长还要看是否有合适的激光器。尺寸、性能、稳定性和成本都得考虑。紫外激光器仍然成本更高而且体积更大。高紫外能量可能损伤样品，而紫外增强CCD相机售价也很贵。所以，很多拉曼光谱仍使用红光或近红外激光(660到830 nm)抑制荧光，即使近红外拉曼散射效率明显更低。图3. Cobolt激光器将所有这些因素考虑在内，拉曼光谱最常用的波长是785 nm——在散射效率、荧光影响、探测效率和可用激光器之间提供最佳平衡。不过，蓝光和绿光(特别是532 nm)用得越来越多，如图4所示。这个范围特别适合无机材料，比如碳纳米管和富勒烯；共振拉曼实验；表面增强拉曼光谱(SERS)。图4. 基于三种不同波长的拉曼谱：对于532和785 nm激光，拉曼信号被荧光淹没；但405 nm激光则很容易分辨。紫外激光很适合蛋白质、DNA和RNA等生物分子的共振拉曼光谱分析。如果荧光很强的材料需要近红外照明，一般使用1064 nm。1064 nm是傅里叶变换拉曼配置的传统选择，它使用单元件铟镓砷或锗探测器避免荧光干涉。不过傅里叶变化拉曼受限于长采集时间以及运动部件要求。由于有了灵敏的铟镓砷阵列器件，如今便携式和手持式拉曼仪器中能够使用1064 nm激光和固定光栅。其它重要的激光参数除了波长，选择拉曼光谱用激光器时还要考虑诸多其它重要参数：光谱线宽、频率稳定性、光谱纯度、光束质量、输出功率和稳定性以及光隔离。最后还要考虑紧凑性、稳固性、可靠性、寿命和成本。拉曼仪器已经成了很多科研和工业应用的标准分析工具。用户希望在运行常规实验或过程监控测量时能够几年不用维修或更换激光器。这些仪器在恶劣工业环境中的运行也越来越多。基于这些原因，现在多数拉曼系统都配备固体而不是气体激光器。紧凑的固体激光器工作寿命长达几万小时，满足最高的光学性能要求，可用波长覆盖所有常用拉曼光谱技术。激光技术用于拉曼光谱仪的连续波固体激光器可分三类：二极管泵浦激光器：单纵模(SLM)单模二极管激光器：分布反馈(DFB)或分布布拉格反射体(DBR)体布拉格光栅(VBG)稳频二极管激光器这三种激光覆盖不同的波长范围，光学性能差异也很大，具体解释如下。集成非线性光学频率转换的二极管泵浦SLM激光器尺寸小，波长范围从紫外到近红外。近红外1064 nm的功率水平可达瓦量级，蓝绿红范围提供很多激光线(660、640、561、532、515、491、473和457 nm)，功率在数百毫瓦量级。紫外功率较低，355 nm激光功率在10到50 mW。这些激光器提供TEM00光束、低漂移精密波长以及线宽一般远小于1 MHz的单频线宽。这种激光光谱纯度很高，主峰数皮米范围内的边模抑制比一般高于60 dB。附近激光线可能以低水平发射，但是偏离主峰数纳米，因此用介质膜滤光片很容易消除。波长稳定性也较高。图5. Cobolt二极管泵浦单纵模激光的波长曲线、稳定性和光束轮廓DFB或DBR单模二极管激光器是紧凑且高性价比光源，提供小于1 MHz的单频线宽和单横模光束。从红光到近红外有很多可选波长，最常用的是785、830、980和1064 nm。边带发射将边模抑制比限制在50 dB左右，一般距主峰几百皮米。对于DFB或DBR光源没有的窄线宽波长，VBG稳频二极管激光器可以使用窄线宽VBG元件获得。它们还能通过多横模二极管激光器的锁频提供更高功率的窄线宽输出。为了稳定输出波长和线宽，尤其是在温度变化的环境下，激光器内部需要精确控制热机械性能和对准。线宽范围从单频到几十皮米，具体取决于波长和功率。和其它二极管激光器类似，主峰附近的边模抑制比为40到50 dB，但是通过带通滤光片能够进一步提高。图6. 红色曲线为加滤光片的结果，黄色曲线不加滤光片拉曼光谱一直是尖端的分析技术，未来还将应用于更多的行业和市场。系统越来越小，分辨率和灵敏度越来越高，组件也越来越可靠。理解激光性能的影响将是拉曼光谱在这些新兴应用领域不断取得成功的关键。Cobolt | Coherence Matters发布于 2020-03-18 20:49激光器光学拉曼光谱赞同 307 条评论分享喜欢收藏申请转载文章被以下专栏收录科技翻译翻译光电技术和应用笔记是一个学习
二极管激光器 - HÜBNER Photonics & Cobolt 激光器
激光器 - HÜBNER Photonics & Cobolt 激光器 跳到内容 APE Singapore 6 - 8 March | Booth: EG-16B, OPIE Japan 24 - 26 April | Booth: B-12, Optatec Germany 14 - 16 May | Booth: 3.1 Booth: 213, Eurosatory France 17 - 21 June | Booth: 3.1 Booth: 213 公司活动使用政策与规范寻找： 产品中心 激光光学 F-Theta 扫描镜头扩束器贝塞尔透镜激光测距仪精密光学“ 查看全部成像光学 短波红外镜头长波红外镜头红外变焦镜头物镜视觉镜片“ 查看全部消费光学 内窥镜镜片非球面镜片监控镜头飞行时间镜头狙击镜“ 查看全部光纤 （合作伙伴的产品）紧凑型光纤耦合器光纤元件“ 查看全部激光器和探测器 （合作伙伴的产品）紧凑型激光器光纤激光器光电二极管光谱仪“ 查看全部系统和软件 （合作伙伴的产品）DTM工具频率梳AOI机分光光度计“ 查看全部应用领域 AR / VR 无人机相机食品和包装检验激光清洗激光切割激光钻孔 激光雕刻激光加工激光扫描激光焊接用于自动驾驶汽车的激光雷达 机器视觉质量温度筛选医疗激光治疗医学光学成像手机相机 扫地机器人监控战术运动“ 查看全部公司能力 定制光学光学设计系统定制自学资料库 作品目录知识中心视频光学计算器光学材料联系我们 菜单切换 产品中心菜单切换激光光学成像光学消费光学光纤激光器和探测器系统和软件应用领域公司能力菜单切换定制光学光学设计系统定制自学资料库菜单切换作品目录知识中心光学计算器光学材料视频联系我们 产品中心 / 激光器和探测器（合作伙伴的产品） / HÜBNER 光子学和 Cobolt 激光器 / 二极管激光器 二极管激光器二极管激光器是一种使用半导体二极管作为增益介质的激光器。 它产生窄波长范围内的相干光，通常从电磁波谱的紫外线到可见光到近红外区域。 二极管激光器的工作原理是让电流穿过半导体材料，通常是镓、砷和其他元素的组合。 这个过程使半导体发射光子，然后由谐振腔放大，产生相干光束。Cobolt 06-01 系列以紧凑的即插即用格式提供大波长范围的激光器。 该系列包括高性能固定波长二极管激光器模块 (MLD) 和二极管泵浦激光器 (DPL)。 高速直接调制能力和调制期间的真正关闭使它们成为生物成像和量子技术应用的理想选择。 Wavelength Opto-Electronic 是的合作伙伴 HÜBNER Photonics 和 Cobolt 品牌的激光器适用于 新加坡, 马来西亚及泰国.v产品中心亮点询价产品波长功率光纤尾纤选项调制速度（上升时间）钴06-MLD375纳米70毫瓦-150 MHz (<2.5 ns)钴06-MLD405纳米150毫瓦365毫瓦75毫瓦150毫瓦150 MHz (<2.5 ns)钴06-MLD415纳米120毫瓦60毫瓦150 MHz (<2.5 ns)钴06-MLD425纳米120毫瓦60毫瓦150 MHz (<2.5 ns)钴06-MLD445纳米100毫瓦400毫瓦50毫瓦150毫瓦150 MHz (<2.5 ns)钴06-MLD457纳米100毫瓦400毫瓦50毫瓦150毫瓦150 MHz (<2.5 ns)钴06-MLD473纳米100毫瓦300毫瓦50毫瓦150毫瓦150 MHz (<2.5 ns)钴06-MLD488纳米60毫瓦100毫瓦150毫瓦200毫瓦30毫瓦100毫瓦150 MHz (<2.5 ns)钴06-MLD505纳米80毫瓦40毫瓦150 MHz (<2.5 ns)钴06-MLD515纳米80毫瓦150毫瓦40毫瓦75毫瓦150 MHz (<2.5 ns)钴06-DPL532纳米25毫瓦50毫瓦100毫瓦200毫瓦-<50 kHz(<6 我们)钴06-DPL553纳米25毫瓦50毫瓦-<5 kHz(<60 我们)钴06-DPL561纳米25毫瓦50毫瓦100毫瓦-<10 kHz(<30 我们)钴06-MLD633纳米80毫瓦40毫瓦150 MHz (<2.5 ns)钴06-MLD638纳米180毫瓦80毫瓦150 MHz (<2.5 ns)钴06-MLD647纳米130毫瓦80毫瓦150 MHz (<2.5 ns)钴06-MLD660纳米100毫瓦50毫瓦150 MHz (<2.5 ns)钴06-MLD685纳米40毫瓦-150 MHz (<2.5 ns)钴06-MLD730纳米50毫瓦-150 MHz (<2.5 ns)钴06-MLD760纳米25毫瓦-150 MHz (<2.5 ns)钴06-MLD785纳米250毫瓦-150 MHz (<2.5 ns)钴06-MLD808纳米120毫瓦-150 MHz (<2.5 ns)钴06-MLD830纳米100毫瓦-150 MHz (<2.5 ns)钴06-MLD940纳米250毫瓦-150 MHz (<2.5 ns)钴06-MLD975纳米120毫瓦-150 MHz (<2.5 ns)375 nm – 975 nm，高达 400 mW完全集成的电子设备所有二极管波长上的集成净化过滤器数字和模拟调制在调制期间真正关闭 (>70 dB)提供光纤尾纤配置兼容 C-FLEX 激光合路器超坚固、久经考验的可靠性产品页面联系表名字组织手机号码邮箱地址我们建议使用您所在组织的电子邮件及其自己的域（如果有）。1) 零件名称/编号数量2) 零件名称/编号数量3) 零件名称/编号数量 添加更多零件名称/编号4) 零件名称/编号数量5) 零件名称/编号数量6) 零件名称/编号数量其它咨询/备注订阅电邮通讯 激光光学 成像光学 消费光学 光纤（合作伙伴的产品） 激光器和探测器（合作伙伴的产品） 系统和软件（合作伙伴的产品）我同意让本网站存储我提交的信息，以便他们回复我的询问提交RFQ更多产品和信息，请点击 点击此处.产品中心 激光光学成像光学消费光学光纤激光器和探测器系统和软件气凝胶新产品按应用搜索产品制造能力关于我们 公司联系我们活动使用政策与规范自学资料库 作品目录知识中心光学计算器光学材料视频订阅 * 必填项电子邮件 *产品分类 激光光学 成像光学 消费光学 光纤/激光器和探测器/系统和软件 下载光学计算器品牌与合作伙伴还有更多品牌可供选择！ISO 9001 认证公司 © 2024 - 波长光电（新加坡）私人有限公司 | 版权所有版权所有
窄线宽激光器 - HÜBNER Photonics & Cobolt Lasers
激光器 - HÜBNER Photonics & Cobolt Lasers 跳到内容 APE Singapore 6 - 8 March | Booth: EG-16B, OPIE Japan 24 - 26 April | Booth: B-12, Optatec Germany 14 - 16 May | Booth: 3.1 Booth: 213, Eurosatory France 17 - 21 June | Booth: 3.1 Booth: 213 公司活动使用政策与规范寻找： 产品中心 激光光学 F-Theta 扫描镜头扩束器贝塞尔透镜激光测距仪精密光学“ 查看全部成像光学 短波红外镜头长波红外镜头红外变焦镜头物镜视觉镜片“ 查看全部消费光学 内窥镜镜片非球面镜片监控镜头飞行时间镜头狙击镜“ 查看全部光纤 （合作伙伴的产品）紧凑型光纤耦合器光纤元件“ 查看全部激光器和探测器 （合作伙伴的产品）紧凑型激光器光纤激光器光电二极管光谱仪“ 查看全部系统和软件 （合作伙伴的产品）DTM工具频率梳AOI机分光光度计“ 查看全部应用领域 AR / VR 无人机相机食品和包装检验激光清洗激光切割激光钻孔 激光雕刻激光加工激光扫描激光焊接用于自动驾驶汽车的激光雷达 机器视觉质量温度筛选医疗激光治疗医学光学成像手机相机 扫地机器人监控战术运动“ 查看全部公司能力 定制光学光学设计系统定制自学资料库 作品目录知识中心视频光学计算器光学材料联系我们 菜单切换 产品中心菜单切换激光光学成像光学消费光学光纤激光器和探测器系统和软件应用领域公司能力菜单切换定制光学光学设计系统定制自学资料库菜单切换作品目录知识中心光学计算器光学材料视频联系我们 产品中心 / 激光器和探测器（合作伙伴的产品） / HÜBNER 光子学和 Cobolt 激光器 / 窄线宽激光器 窄线宽激光器Cobolt 08-01 系列是紧凑型窄线宽连续波激光器系列，包括高性能单频二极管泵浦激光器 (DPL) 和稳频二极管激光器 (NLD)。 该系列为拉曼光谱提供市场上最宽的波长范围和最高性能的激光器。 Wavelength Opto-Electronic 是 HÜBNER Photonics 和 Cobolt 品牌激光器的合作伙伴 新加坡, 马来西亚及泰国。产品中心亮点询价产品波长功率带宽钴08-NLD405纳米30毫瓦40毫瓦< 下午 1 点钴08-DPL457纳米25毫瓦30毫瓦< 1 兆赫钴08-DPL473纳米25毫瓦40毫瓦50毫瓦< 1 兆赫钴08-NLD488纳米40毫瓦< 下午 1 点钴08-DPL515纳米25毫瓦50毫瓦< 1 兆赫钴08-DPL532纳米25毫瓦50毫瓦100毫瓦200毫瓦< 1 兆赫钴08-DPL561纳米25毫瓦50毫瓦100毫瓦< 1 兆赫钴08-NLD633纳米30毫瓦< 下午 1 点钴08-NLD638纳米80毫瓦< 下午 1 点钴08-DPL660纳米50毫瓦< 1 兆赫钴08-NLD785纳米120毫瓦< 下午 1 点钴08-NLDM785纳米500毫瓦< 下午 70 点钴螺栓 08-NLDM-ESP785纳米400毫瓦< 下午 70 点钴08-NLD830 纳米 新！100毫瓦< 下午 1 点钴08-DPL1064纳米400毫瓦< 1 兆赫可用波长：375 nm 至 1064 nm，最高 1000 mW易于安装和现场升级高速调制具有多个输出选项的光纤耦合可选机电光圈快门产品页面联系表名字组织手机号码邮箱地址我们建议使用您所在组织的电子邮件及其自己的域（如果有）。1) 零件名称/编号数量2) 零件名称/编号数量3) 零件名称/编号数量 添加更多零件名称/编号4) 零件名称/编号数量5) 零件名称/编号数量6) 零件名称/编号数量其它咨询/备注订阅电邮通讯 激光光学 成像光学 消费光学 光纤（合作伙伴的产品） 激光器和探测器（合作伙伴的产品） 系统和软件（合作伙伴的产品）我同意让本网站存储我提交的信息，以便他们回复我的询问提交RFQ更多产品和信息，请点击 点击此处.产品中心 激光光学成像光学消费光学光纤激光器和探测器系统和软件气凝胶新产品按应用搜索产品制造能力关于我们 公司联系我们活动使用政策与规范自学资料库 作品目录知识中心光学计算器光学材料视频订阅 * 必填项电子邮件 *产品分类 激光光学 成像光学 消费光学 光纤/激光器和探测器/系统和软件 下载光学计算器品牌与合作伙伴还有更多品牌可供选择！ISO 9001 认证公司 © 2024 - 波长光电（新加坡）私人有限公司 | 版权所有版权所有
可调谐激光器 - HÜBNER Photonics & Cobolt Lasers
激光器 - HÜBNER Photonics & Cobolt Lasers 跳到内容 APE Singapore 6 - 8 March | Booth: EG-16B, OPIE Japan 24 - 26 April | Booth: B-12, Optatec Germany 14 - 16 May | Booth: 3.1 Booth: 213, Eurosatory France 17 - 21 June | Booth: 3.1 Booth: 213 公司活动使用政策与规范寻找： 产品中心 激光光学 F-Theta 扫描镜头扩束器贝塞尔透镜激光测距仪精密光学“ 查看全部成像光学 短波红外镜头长波红外镜头红外变焦镜头物镜视觉镜片“ 查看全部消费光学 内窥镜镜片非球面镜片监控镜头飞行时间镜头狙击镜“ 查看全部光纤 （合作伙伴的产品）紧凑型光纤耦合器光纤元件“ 查看全部激光器和探测器 （合作伙伴的产品）紧凑型激光器光纤激光器光电二极管光谱仪“ 查看全部系统和软件 （合作伙伴的产品）DTM工具频率梳AOI机分光光度计“ 查看全部应用领域 AR / VR 无人机相机食品和包装检验激光清洗激光切割激光钻孔 激光雕刻激光加工激光扫描激光焊接用于自动驾驶汽车的激光雷达 机器视觉质量温度筛选医疗激光治疗医学光学成像手机相机 扫地机器人监控战术运动“ 查看全部公司能力 定制光学光学设计系统定制自学资料库 作品目录知识中心视频光学计算器光学材料联系我们 菜单切换 产品中心菜单切换激光光学成像光学消费光学光纤激光器和探测器系统和软件应用领域公司能力菜单切换定制光学光学设计系统定制自学资料库菜单切换作品目录知识中心光学计算器光学材料视频联系我们 产品中心 / 激光器和探测器（合作伙伴的产品） / HÜBNER 光子学和 Cobolt 激光器 / 可调谐激光器C波钴蓝奥丁可调谐激光器我们提供两种类型的可调谐激光器； CW & Q 切换。 所有可调谐激光器均基于 OPO 技术，涵盖 VIS – MIR 波长范围。 Wavelength Opto-Electronic 是 HÜBNER Photonics 和 Cobolt 品牌激光器的合作伙伴 新加坡, 马来西亚及泰国。产品中心资讯询价C波钴奥丁™C-WAVE模型调音范围输出功率C-WAVE 可见光 低电量450 - 525 纳米540 - 650 纳米900 - 1050 纳米1080 - 1300 纳米高达 200 毫瓦高达 200 毫瓦高达 400 毫瓦高达 400 毫瓦C-WAVE 可见光 高功率450 - 525 纳米540 - 650 纳米900 - 1050 纳米1080 - 1300 纳米高达 500 毫瓦高达 500 毫瓦最高为1 W.最高为1 W.C波GTR500 - 750 纳米1000 - 1500 纳米1700 - 3500 纳米最高为1 W.最高为1.5 W.最高为2 W.C波红外 低电量900 - 1050 纳米1080 - 1300 纳米高达 400 毫瓦高达 400 毫瓦C波红外 高功率900 - 1050 纳米1080 - 1300 纳米最高为1 W.最高为1 W.C-波近红外1000 - 1500 纳米1700 - 3500 纳米最高为1.5 W.最高为2 W.产品波长功率钴奥丁™3264纳米*> 80 mW @ 10 kHz钴奥丁™3431纳米*> 80 mW @ 10 kHz钴奥丁™4330纳米*> 60 mW @ 10 kHz波钴奥丁™C-WAVE 可调谐激光器是一种基于光学参量振荡器 (OPO) 技术的广泛可调连续波 (cw) 单频激光光源。 它在可见光和近红外波长范围内提供完全由计算机控制的发射调谐。波长覆盖范围从 450 nm 到 3.5 µm高达 250 nm 的可见光无间隙调谐范围 单频操作，典型线宽 < 500 kHz输出功率高达瓦特级Cobolt Odin™ 系列是一种超紧凑型工业级中红外光源，基于带有集成泵浦激光器的全封闭温度可调光学参量振荡器 (OPO)。波长可选 3,1 µm – 4,6 µm标准波长 3264 nm、3431 nm 和 4330 nm可调至 50 nm80 kHz 时高达 10 mW窄带宽 <0.3 nm超坚固、密封封装、久经考验的可靠性 产品页面联系表名字组织手机号码邮箱地址我们建议使用您所在组织的电子邮件及其自己的域（如果有）。1) 零件名称/编号数量2) 零件名称/编号数量3) 零件名称/编号数量 添加更多零件名称/编号4) 零件名称/编号数量5) 零件名称/编号数量6) 零件名称/编号数量其它咨询/备注订阅电邮通讯 激光光学 成像光学 消费光学 光纤（合作伙伴的产品） 激光器和探测器（合作伙伴的产品） 系统和软件（合作伙伴的产品）我同意让本网站存储我提交的信息，以便他们回复我的询问提交RFQ更多产品和信息，请点击 点击此处.产品中心 激光光学成像光学消费光学光纤激光器和探测器系统和软件气凝胶新产品按应用搜索产品制造能力关于我们 公司联系我们活动使用政策与规范自学资料库 作品目录知识中心光学计算器光学材料视频订阅 * 必填项电子邮件 *产品分类 激光光学 成像光学 消费光学 光纤/激光器和探测器/系统和软件 下载光学计算器品牌与合作伙伴还有更多品牌可供选择！ISO 9001 认证公司 © 2024 - 波长光电（新加坡）私人有限公司 | 版权所有版权所有
Cobolt: integrative analysis of multimodal single-cell sequencing data | Genome Biology | Full Text
Cobolt: integrative analysis of multimodal single-cell sequencing data | Genome Biology | Full Text
Skip to main content
Advertisement
Search
Explore journals
Get published
About BMC
My account
Search all BMC articles
Search
Genome Biology
Home
About
Articles
Submission Guidelines
Submit manuscript
Cobolt: integrative analysis of multimodal single-cell sequencing data
Download PDF
Download PDF
Method
Open access
Published: 28 December 2021
Cobolt: integrative analysis of multimodal single-cell sequencing data
Boying Gong1, Yun Zhou1 & Elizabeth Purdom
ORCID: orcid.org/0000-0001-9455-79902
Genome Biology
volume 22, Article number: 351 (2021)
Cite this article
12k Accesses
33 Citations
28 Altmetric
Metrics details
AbstractA growing number of single-cell sequencing platforms enable joint profiling of multiple omics from the same cells. We present Cobolt, a novel method that not only allows for analyzing the data from joint-modality platforms, but provides a coherent framework for the integration of multiple datasets measured on different modalities. We demonstrate its performance on multi-modality data of gene expression and chromatin accessibility and illustrate the integration abilities of Cobolt by jointly analyzing this multi-modality data with single-cell RNA-seq and ATAC-seq datasets.
BackgroundSingle-cell sequencing allows for quantifying molecular traits at the single-cell level, and there exist a wide variety of platforms that extend traditional bulk platforms, such as mRNA-seq and ATAC-seq, to the single cell. Comparison of different cellular features or modalities from cells from the same biological system gives the potential for a holistic understanding of the system. Most single-cell technologies require different cells as input to the platform, and therefore, there remains the challenge of linking together the biological signal from the different modalities, with several computational methods proposed to estimate the linkage between the different modalities, such as LIGER [1] and Signac (Seurat) [2, 3].Recently, there are a growing number of platforms that allow for measuring several modalities on a single cell. CITE-seq [4] jointly sequences epitope and transcriptome; scNMT-seq [5] jointly profiles chromatin accessibility, DNA methylation, and gene expression; and sci-CAR [6], Paired-seq [7], and SNARE-seq [8] enable simultaneous measurement of transcription and chromatin accessibility (we direct readers to [9] for a comprehensive review). By directly measuring the different modalities on the same cells, these techniques greatly enhance the ability to relate the different modalities. With the emergence of joint platforms, new computational methodologies for analyzing multi-modality data have also been developed. Early methods mainly focused on CITE-seq [10–12], which jointly sequences gene expressions and at most a few hundred antibodies. Recently, more methods have been proposed to enable the modeling of cells with simultaneous measurement of gene expression and higher-dimensional modalities such as chromatin accessibility, such as MOFA+ (also known as MOFA2 [13]), scMVAE [14], BABEL [15], and scMM [16].However, single-cell datasets on a single modality are far more common and are usually of higher throughput. Indeed, it is natural that joint-modality data from the same system will be used to augment single-modality data, or vice versa. Therefore, there is a critical need for an analysis tool that is both a stand-alone application for multi-modality data as well as a tool for integration of these datasets with single-modality platforms. BABEL [15] and scMM [16], while not directly targeting this task, do allow the use of the joint-modality data to predict one single-modality dataset into another type of modality. However, neither directly integrate the data together to allow for downstream analysis of the joint set of data regardless of modality, such as cell subtype detection.Our method Cobolt fills this gap by providing a coherent framework for a full integrative analysis of multi-modality and single-modality platforms. The result of Cobolt is a single representation of the cells irrespective of modalities, which can then be used directly by downstream analyses, such as joint clustering of cells across modalities. Cobolt estimates this joint representation via a novel application of Multimodal Variational Autoencoder (MVAE) [17] to a hierarchical generative model. The integration of the single-modalities is done by a transfer learning approach which harnesses the valuable information found by joint sequencing of the same cells and extends it to the cells in the single-cell platform. The end result is a single representation of all of the input cells, whether sequenced on a multi-modality platform or a single-modality platform. In this context, Cobolt gives an over-all integrative framework that is flexible for a wide range of modalities.We demonstrate Cobolt on two use-cases. The first uses Cobolt to analyze only a multi-modality sequencing dataset from the SNARE-seq technology; we show that Cobolt provides a joint analysis that better distinguishes important facets of each modality, compared to existing methods. The second demonstrates the use of Cobolt to integrate multi-modality data with single-modality data collected from related biological systems, where Cobolt creates a joint representation that can be used for downstream analysis to provide meaningful biological insights. We show that Cobolt also performs better than related tools in this integrative task.ResultsThe Cobolt modelWe develop a novel method, Cobolt, that utilizes joint-modality data to enable the joint analysis of cells sequenced on separate sequencing modalities. We do this by developing a Multimodal Variational Autoencoder based on a hierarchical Bayesian generative model. We briefly describe the premise of the model using the example of two modalities: mRNA-seq and ATAC-seq (for more details in greater generality, see the “Methods” section). We assume that we have a set of cells with both mRNA-seq and ATAC-seq data collected using the joint-modality platform (\(X_{1}^{mRNA}\) and \(X_{1}^{ATAC}\)), as well as (optionally) a set of cells with only mRNA-seq data (\(X_{2}^{mRNA}\)), and a set of cells with only ATAC-seq data (\(X_{3}^{ATAC}\)). Cobolt takes all of this data as input to find a representation of the cells in a shared reduced dimensionality space, regardless of modality (Fig. 1A).
Fig. 1An overview of the Cobolt method. The upper panel shows the workflow of Cobolt, which takes as input datasets with varying modalities, projects the data into shared latent space, and then performs visualization and clustering with all datasets combined. The lower panel shows the Cobolt variational autoencoder model with encoders plotted on the left and decoders on the rightFull size imageCobolt models the sequence counts from the modalities inspired by the Latent Dirichlet Allocation (LDA) [18] model. The LDA model is a popular Bayesian model for count data which has been successfully applied to genomics in areas such as clustering of single-cell RNA-seq data [19, 20], single-cell ATAC-seq analysis [21], and functional annotation [22]. Cobolt builds a hierarchical latent model to model data from different modalities and then adapts the MVAE approach to both estimate the model and allow for a transfer of learning between the joint-modality data and the single-modality data.Cobolt assumes that there are K different types of possible categories that make up the workings of a cell. For ease of understanding, it is useful to think of these categories as biological processes of the cell, though the categories are unlikely to actually have a one-to-one mapping with biological processes. Each category will result in different distributions of features in each of the modalities—i.e., different regions of open chromatin in ATAC-seq or expression levels of genes in mRNA-seq for different categories. The features measured in a cell are then the cumulative contribution of the degree of activation of each category present in that cell. The activation level of each category is represented by the latent variable θc for each cell c, which gives the relative activity of each of the K categories in the cell. θc is assumed to be an intrinsic property of each cell representing the underlying biological properties of the cell, while the differences of data observed in each modality for the same cell are due to the fact that the categories active in a cell have different impacts in the modality measured (open chromatin in ATAC-seq versus gene expression in mRNA-seq). We assume θc=σ(zc), where zc is drawn from a Gaussian prior and σ is the soft-max transformation; this is an approximation to the standard Dirichlet prior for θc that allows use of variational autoencoders to fit the model [23]. The mean of the posterior distribution gives us an estimate of our latent variable zc for each cell, and the posterior distribution is estimated using variational autoencoders (VAE).In the end, Cobolt results in an estimate of the latent variable zc for each cell, which is a vector that lies in a K-dimensional space. This space represents the shared biological signal of the individual cells, irregardless of modality, and can be used for the common analysis tasks of single-cell data, such as visualization and clustering of cells to find subtypes. Importantly, we can predict the latent variable zc even when a cell does not have all modalities measured. Moreover, because of the joint-modality platforms, Cobolt does not require that the different modalities measure the same features in order to link the modalities together—the fact that some of the cells were sequenced on both platforms provides the link between different types of features. Therefore, ATAC-seq peaks and mRNA-seq gene expression can be directly provided as input. This is unlike methods that do not make use of the joint-modality data and require that the different modalities be summarized on the same set of features, for example by simplifying ATAC-seq peaks to a single measurement per gene.
Cobolt as an analysis tool for multi-modality dataWhile the full power of Cobolt is to integrate together data from single and multiple modality datasets, in its simplest form, Cobolt can be used for the analysis of data from solely a multi-modality technology. We demonstrate this usage with the SNARE-seq data [8], which consists of paired transcription and chromatin accessibility sequenced on 10,309 cells of adult mouse cerebral cortices.We first compare Cobolt with a simple, but common, approach for analyzing joint-modality data: the two modalities are analyzed separately and then the results are linked together. This is the strategy of [8], where the authors primarily clustered the gene expression modality to form clusters of the cells, and then performed a separate analysis on chromatin accessibility modality as a comparison. Focusing on the gene expression modality is common, since it is often assumed to have the greatest resolution in determining cell types. However, the reduced representations and clusters created on one modality may not be representative of all the underlying cell subtypes. Indeed, when we perform clustering analysis on only the gene expression modality using Seurat [3] and only the chromatin accessibility modality using cisTopic [21] (consistent with [8], see the “Methods” section), both modalities find distinct clusters that are not reflected in the other modality (Additional file 1: Fig. S1). For example, cells identified by marker genes as non-neuronal cells, such as astrocytes, oligodendrocytes, oligodendrocyte precursors, and microglial cells, are clustered into their respective cell types based on mRNA expression but are not separated based only on chromatin accessibility (Additional file 1: Fig. S2A). Similarly, a subset of layer 5/6 cells has distinct chromatin accessibility peaks but are intermingled with other layer 5/6 cells in the gene expression clusters; these differential peaks include one near the gene Car12 which is a marker gene of the previously annotated subtype of layer 6 (L6 Car12, [24]) and which shows higher expression in this subset of cells (Additional file 1: Fig. S2B,C). A joint analysis with Cobolt, unlike the single-modality analyses, finds these subtypes detected by only one modality and not the other (Additional file 1: Fig. S1).Next, we compare with other methods that explicitly analyze the two modalities jointly, like Cobolt. We consider the methods MOFA2, scMM, and BABEL. MOFA2 uses Bayesian group factor analysis for dimensionality reduction of multi-modality datasets; BABEL trains an interoperable neural network model on the paired data that translates data from one modality to the other; and scMM [16] uses a deep generative model for joint representation learning and cross-modal generation. We apply each of these methods to the SNARE-seq data.Since a joint analysis method should be able to reflect subtype signals captured by all modalities, we similarly evaluate the methods on how well their lower-dimensional spaces represent separate clusters identified separately in each modality, as described above. In Fig. 2, we visualize the lower-dimensional space generated by Cobolt, MOFA2, scMM, and BABEL via UMAP (Uniform Manifold Approximation and Projection [25]), where we color the cells based on the clusters found by clustering the gene expression modality data (Fig. 2A) and those based on clustering the chromatin accessibility data (Fig. 2B). We would note that BABEL does not create a single reduced-dimensionality representation for a paired cell, but rather one per modality (the two latent representations are learned jointly and are quite similar). BABEL’s lower-dimensionality representation does quite poorly, separating major clusters of cells found in both modalities, such as layer 2 to 6 intratelencephalic (IT) neurons (colored red, purple, pink, and cyan in Fig. 2A). Both MOFA2 and scMM capture these large clusters, which are shared between the modalities. However, we see clusters specific to a single modality not reflected on their lower-dimensional space. For example, the highlighted gene expression cluster in Fig. 2A is practicably indistinguishable in the scMM UMAP but separated in the Cobolt UMAP. Differential mRNA expression analysis between this cluster and neighboring cells finds strong expression of known markers of layer 6 cells in this cluster (Col24a1 [26], Gnb4 [27], Rxfp1 [28], Nr4a2 [29, 30], and Ntng2 [29], Additional file 1: Fig. S4A) as well as strong expression of Car3 defined in [31] as a marker of a subset of layer 6 IT cells. Neighboring cells do not express these known marker genes and instead express layer 5/6 IT markers (cyan cluster) or layer 2/3 IT markers (red cluster), indicating that this cluster missed by scMM consists of a biologically meaningful subset of layer 6 IT cells. Similarly, in Fig. 2B, we highlight two clusters of cells which clearly separate in the Cobolt analysis and are separate clusters based on both mRNA-Seq and chromatin profiles, but are mixed together in the MOFA2 analysis. Differential mRNA expression analysis between these clusters reveal genes Adarb2 and Sox6 differ in expression between these groups (Additional file 1: Fig. S4B), which are known markers whose expression distinguish the CGE and Pvalb clusters, respectively [31]. Integrative analysis in the next section confirms this identification by integrating this SNARE-Seq data with annotated scRNA-Seq data and placing these cells with cells annotated as CGE and Pvalb in [31].
Fig. 2Comparison of multi-modality analysis methods. A, B UMAP visualizations of the reduced dimensionality space created by Cobolt, MOFA2, scMM, and BABEL. The cells are color-coded by the cluster they are assigned to based on clustering of A only gene expression modality and B only chromatin accessibility modality. We note that the cluster colors are randomly and separately assigned for panels A and B. Highlighted in the panels are clusters that are well separated in the analysis of Cobolt, but not the other methods. More details on the silhouettes per cluster can be found in Additional file 1: Fig. S3Full size imageTo quantitatively evaluate these observations, we calculate the average silhouette widths of the modality-specific clustering on the UMAPs generated by Cobolt, MOFA2, scMM, and BABEL (shown in Fig. 2C), where higher silhouette widths indicate that cells are closer to other cells in the same cluster. As expected from our observations, BABEL’s representation results in extremely small silhouette widths, reflecting the many clusters separated in BABEL’s representation. MOFA2 has the smallest silhouette width on chromatin accessibility clusters, supporting our observation that its joint space does not represent this modality well; similarly, scMM gives relatively small measures on the gene expression modality. Cobolt best represents both modalities with the highest silhouette width measure.
Cobolt for integrating multi-modality data with single-modality dataWe now turn to integrating multi-modality data with single-modality data. For this use-case, we use Cobolt to jointly model three different datasets—the SNARE-seq of mouse cerebral cortices analyzed in the above section, together with a scRNA-seq and a scATAC-seq dataset of mouse primary motor cortex (MOp) [31]. In addition, we also demonstrate Cobolt for joint modeling of single-cell sequencing of human peripheral blood mononuclear cells (PBMCs): two multi-modality datasets pairing ATAC and mRNA measurements on 10,970 and 12,012 cells from different samples of the 10X Multiome platform [32, 33], combined with 23,837 cells from scRNA-Seq [34] and 9030 cells from scATAC-Seq [35].The result in both examples is a lower-dimensional latent space that aligns the different modality data into a single representation. In Fig. 3A and B, we visualize this low-dimensional space via UMAP for the mouse cortex and human PBMCs, respectively, with cells colored by their data set of origin. We see that the cells from different datasets are well aligned regardless of their source of origin.
Fig. 3The UMAP visualization of A, B the mouse cortex integration and C, D the 10X PBMC integration. Cells are colored in A, C by dataset of origin, in B known cell type annotation of [31], and in D by our de novo clustering and annotated based on gene markers. For the mouse cortex integration, both the MOp scRNA-seq and the MOp scATAC-seq contain a substantial fraction of cells labeled “unannotated” by the authors of the data and that do not map to known cell types. The cell type abbreviation largely follows the data paper [31]: astrocytes (Astro), caudal ganglionic eminence interneurons (CGE), endothelial cells (Endo), layer 2 to layer 6 (L2-6), intratelencephalic neurons (IT), pyramidal tracts (PT), corticothalamic neurons (CT), L6b excitatory neurons (L6b), microglial cells (MGC), near-projecting excitatory neurons (NP), oligodendrocytes (Oligo), oligodendrocyte precursors (OPC), smooth muscle cells (SMC), and medial ganglionic eminence interneurons subclasses based on marker genes (Sst, Pvalb). For the 10X PBMC integration, the following abbreviations are observed: dendritic cell (DC), plasmacytoid dendritic cells (pDC), monocytes (mono), and natural killer cells (NK)Full size imageTo consider further the biological meaning of the lower-dimensional representation, we label the MOp cells from the mouse cortex dataset in Fig. 3C by their cellular subtype as annotated in [31]. For the purposes of comparison across the modalities, we integrated some cell types in [31] into larger groupings and modified the names so as to have comparable groups (see the “Methods” section). For the SNARE-seq cells, we do not have the cell types given in [8], so we use the identifications found by our analysis of only the SNARE-seq cells (see the “Methods” section). We see that cells from the same cellular subtypes are projected closely regardless of the data source. We also see that the representation of Cobolt respects the larger category of cell types by grouping three major cell classes: GABAergic inhibitory neurons (CGE, Sst, Pvalb), glutamatergic excitatory neurons (IT, L5 PT, L6 CT, L6b, NP), and non-neurons.The PBMC datasets do not have accompanying annotation, so we applied the Louvain clustering algorithm to the lower-dimensional representation from Cobolt to identify potential cell types. Using marker genes [3, 36–38], we classified the clusters into known subtypes expected for PBMC data (Additional file 1: Figs. S5 and S6) and we see that important cell types and functions are localized in the Cobolt representation (Fig. 3C).The mouse cortex data also demonstrates the ability of our joint representation to capture subtype signals that are not shared across all of the modalities. Indeed, despite detecting mostly similar cell types, the MOp datasets profile several cell types distinct to the modality. For example, microglial cells (MGC) and smooth muscle cells (SMC) are uniquely detected in scATAC-seq. The different datasets also have different cellular compositions of their shared subtypes, where astrocytes (Astro) and oligodendrocytes (Oligo) are much more abundant in the scATAC-seq (6.55% and 10.50%) than in the SNARE-seq (4.5% and 2.84%) and scRNA-seq (0.40% and 0.36%). As shown in Fig. 3B, cell populations unique to one dataset are grouped in the UMAP plot and are distinguishable from the other datasets/cell types. This indicates that Cobolt reconciles data even when one cell population is entirely absent or scarcely represented in one or more data sources, which is important in integrating datasets collected from related but slightly differing settings.Cobolt also facilitates subtype identification at a finer resolution by transferring information between modalities. For example, for the mouse cortex data, Fig. 3B shows the cell types based on the published annotation. To make the annotation consistent between scRNA-Seq and scATAC-Seq, some of these cell types are the result of merging some cell types into larger categories (see the “Methods” section). One cell type, caudal ganglionic eminence interneurons (CGE, dark green in Fig. 3B), was annotated in the scATAC-seq MOp dataset as one cluster, but the scRNA-seq annotation further divided CGE cells into 3 subtypes based on marker genes—Lamp5, Vip, and Sncg. Our joint mapping of the cells allows us to relate the subtypes detected in scRNA-seq to scATAC-seq and provides a finer resolution breakdown of CGE in scATAC-seq. Specifically, we ran a de novo clustering of our joint mapping of all three datasets by Cobolt (see the “Methods” section and Additional file 1: Fig. S7). This clustering results in a cluster (cluster 13) composed of Lamp 5 and Sncg cells, while another (cluster 16) is mostly Vip cells (Fig. 4A). This subdivision is further validated by gene expression and gene activity levels in these clusters of marker genes Lamp5 and Vip as well as other genes known to discriminate subtype Lamp5/Sncg from subtype Vip [39], such as Reln and Npy (Fig. 4B). We further validated the scATAC-seq clusters through de novo differential accessibility (DA) and differential expression (DE) analysis (Additional file 1: Fig. S8). We identified 94 DA genes between these two clusters. Seventy-eight of the DA genes are also found DE in the mRNA, and all of them have the same direction of fold changes in the DE and the DA analyses, i.e., the genes with lower/higher gene expression in cluster 13 compared to cluster 16 are also less/more accessible. This shows that the joint model of Cobolt can help distinguish noisy cells in one dataset with additional information from other datasets or modalities.
Fig. 4A We show the relative composition of cells annotated as CGE by the scATAC-seq dataset in the clusters found by clustering of the reduced dimensionality of Cobolt and compare that to the relative composition of the cells annotated in the subtypes Lamp5, Vip, and Sncg by the scRNA-seq dataset. B Plots of the gene expression (scRNA-seq) and gene body accessibility summaries (scATAC-seq) in clusters 13 and 16 of the marker genes that distinguish between cell types Lamp5 and Vip [31, 39]Full size imageFurthermore, Cobolt is robust to poor-quality cells and low-expressed genes by using a count-based model, which should naturally down-weight the influence of low-count cells. In the above analysis of the mouse cortex data, we filtered out 2.5% of MOp mRNA cells and 15.8% ATAC-seq cells due to low counts and other quality control measures following the work of [31]. Yet Cobolt is robust to these choices, even in the extreme case where there is no filtering on cells and only a minimal filter on genes is performed (Additional file 1: Fig. S9).Comparison with existing methodsAs described in the introduction, there are few existing methods that allow full integration of multi-modality sequencing data with single-modality data. MOFA2 only analyzes joint-modality data; BABEL and scMM train a joint model on the paired joint-modality data and allow the user to apply this model to single-modality datasets to predict the other “missing” modality. Unlike Cobolt, these two methods do not use the single-modality data in the training of the model, nor do they provide a representation of the single-modality data in a single shared representation space regardless of modality—for example, for the purposes of joint clustering of all of the cells across modalities.Therefore, to have additional points of comparison, we apply the LIGER and Signac methods, which are designed for integrating unpaired modalities (the implementation details can be found in the “Methods” section). LIGER applies an integrative nonnegative matrix factorization (iNMF) approach to project the data onto a lower-dimensional space and then builds a shared nearest neighbor graph for joint clustering. Signac implements canonical correlation analysis (CCA) for dimensionality reduction; Signac subsequently transfers cell labels by identifying mutual nearest neighbor cell pairs across modalities.To evaluate the performance of these methods when integrating single-modality data with multiple-modality data, we return to the multi-modality datasets described above to create artificial sets of multiple-modality data and single-modality data. For the 10X Multiome data, we make use of the fact that we have two datasets from the 10X Multiome platform run on different patient samples: PBMC of a healthy male donor aged 30–35 (“Multiome Chromium X”) [32] and PBMC of a female donor aged 25 (“Multiome unsorted”) [33]. We ignore the pairing information in the Multiome unsorted data and treat the mRNA and ATAC measurements as coming from unpaired, separate sequencing experiments. For the SNARE-Seq data, we randomly assign the cells to be considered as from either the multi-modality dataset (20%) or the single-modality datasets (80%) and run each of the methods. The choice of 20% and 80% was based on the relative size of the SNARE-Seq joint-modality data to the individual MOp scATAC-Seq data and scRNA-Seq datasets and reflects the fact that single-modality datasets are much higher-throughput than paired-modality data.We give to each of the methods the multi-modality data from the cells that remain paired and for the cells where we ignore the pairing information give the mRNA and ATAC data from those cells as if they were single-modality datasets. For LIGER and Signac, which are not designed for multiple-modality data, we hide the pairing information on all cells and treat all of the cells as if they were collected on different cells. In this way, we have a ground truth on how the cells in the single-modality datasets should be connected to each other, and we can compare the methods by evaluating whether coordinates of the reduced dimensions Z for the pairs of cells assigned to the single-modality datasets were close together. Specifically, for each method, we evaluate for each cell in the mRNA single-modality set its coordinates \(\hat {Z}^{mRNA}\) and for a fixed number k locate its k nearest neighbors in the ATAC single-modality set based on the coordinates \(\hat {Z}^{ATAC}\). We then calculate the percentage of mRNA single-modality cells whose paired cell in the ATAC single-modality is included in its set of nearest neighbors. The reverse analysis was done using chromatin accessibility as the query and evaluating the percentage whose nearest neighbors include their mRNA pair. Many popular clustering routines use nearest-neighbor graphs for identifying clusters, so this is a metric directly related to whether the cells assigned to the single-modality data would likely correctly cluster together across modalities, but avoids having to specify cluster parameters, especially as applied to different methods (and LIGER and Signac have their own clustering techniques specific for their methods).As shown in Fig. 5, the Cobolt joint representation does a much better job for both the SNARE-Seq and 10X multiome data of assigning coordinates to the single-modality cells that place them close to their matching pair. The proportion of single-modality cells that are neighbors to their pair is much larger than any of these other methods in both datasets. Surprisingly, for the SNARE-seq data, the other methods that make use of the joint-modality data to develop their model (scMM and BABEL) do much worse than Signac and LIGER which do not have any information linking the cells together. The 10X multiome data, which has similar numbers of cells from the single-modality datasets as the multi-modality dataset, shows scMM and BABEL perform comparably to Signac, though not as well as Cobolt; similarly, we see improved performance of scMM and BABEL in the SNARE-Seq data when we increase the proportion of dual-modality cells to 80% and only 20% of cells being single-modality (Additional file 1: Fig. S12), but still well below the performance of Cobolt. This points to the power of truly integrating the high-throughput single-modality data into the analysis, particularly when there are more cells sequenced from the single-modality data, as is frequently the case.
Fig. 5Evaluation of dimensionality reduction. A comparison of lower-dimensional representation generated by LIGER, Signac, scMM, BABEL, and Cobolt on A a SNARE-seq data and B two 10X multiome datasets. The x-axis shows the number of neighbors considered (k) as a proportion of the total testing sample size. The y-axis shows the proportion of cells whose paired data are within their k-nearest neighbors in the other modality. The plot gives the results when using gene expression data as the query. We observe very similar results when chromatin accessibility is used as the query (Additional file 1: Fig. S10)Full size imageSubtype detection is a critical aspect of single-cell data analysis, and integrating single-modality data with multiple-modality data gives the potential for higher resolution detection. Our nearest-neighbor metric is directly related to subtype detection, with a higher neighbor proportion corresponding roughly to finding larger clusters in a clustering algorithm. scMM and BABEL only provide cross-modality predictions, rather than a joint embedding of the single-modality data with the multiple-modality data. Downstream tasks such as clustering for subtype detection must be done on either the mRNA expression space or the chromatin expression space, and as we have seen in the SNARE-seq only analysis each of which can miss important features of the data. Indeed, this may be an important factor in their low performance on our nearest-neighbor analysis. LIGER and Signac do provide a joint embedding of all of the data, but do so without the use of the pairing information for the joint-modality data for training the embedding.The previous analysis provides a comparison with a known ground-truth and the evaluation at different levels of analysis. Now, we return to the joint embedding of MOp scRNA-Seq and scATAC-Seq with the SNARE-Seq data that we considered in the previous section and consider the performance of LIGER and Signac, which performed comparatively well. We would note that LIGER and Signac have their own clustering strategies, separate from their dimensionality reduction, but we focus here on the results of their dimensionality reduction. We provided LIGER and Signac only the MOp data, as there is no clear way of including the SNARE-seq into LIGER and Signac without focusing on one of its modalities and adding extra batch correction steps. We compare the results to the clusters published with the scRNA-seq and scATAC-seq data in [31] (Additional file 1: Fig. S13). We see that Cobolt generates a UMAP visualization that well represents rare subpopulations and respects broader cell classes. LIGER gives greater separation between cell types but splits several subtypes into far-away islands, such as for L5 PT, MGC, and Astro. Signac adopts an asymmetric strategy of transferring scRNA-seq labels to scATAC-seq data, and as a result, Signac performs well on major cell types but poorly on under-represented subpopulations in scRNA-seq such as astrocytes (Astro), which accounts for only 0.4% of the cells in scRNA-seq but 6.55% in scATAC-seq. Furthermore, Cobolt, unlike LIGER and Signac, not only groups together the subtypes, but appears to also represent the three broader categories major GABAergic inhibitory neurons, glutamatergic excitatory neurons, and non-neurons.These cluster identifications that we highlight from [31] are relatively robust, well-known cell types, representing the large structural changes in the data, for which we expect most strategies to be able to detect reasonably well. On the other hand, our nearest-neighbor analysis emphasizes the performance at a high level of resolution. Putting both of these together points to the fact that Cobolt provides a superior integration of the datasets across a wide spectrum of resolutions.DiscussionIn this paper, we have shown that Cobolt successfully integrates multi-modal data and provides a representation that can be used for downstream analysis tasks, such as cell-type discovery. Pseudo-time estimation for reconstructing developmental order of cells [40], while not meaningful for the datasets we considered, is another important downstream application where the integrated representation of Cobolt allows the analysis of cells from different modalities. Future work could make use of the graphical model and inferred parameters to establish connections between features. For example, the probability vectors generated by B(i) naturally provide a reduced-dimensional space of molecular features and can potentially help in the construction of gene networks.We would note that while we have focused on the capability of Cobolt to analyze data from two modalities, the underlying method can be extended to larger number of modalities and integration of different combinations of modalities, such as datasets with different pairs of modalities (see the “Methods” section). Thus, Cobolt provides a framework to integrate a wide range of varieties of multi-modality platforms as well as single-modality platforms.Cobolt is available as a Python package at https://github.com/epurdom/cobolt_manuscript. All of the code used for the analysis is available as a github repository: https://github.com/epurdom/cobolt_manuscript.ConclusionsWe have shown that Cobolt is a flexible tool for analyzing multi-modality sequencing data, whether separately or integrated with single-modality data. Cobolt synthesizes the varied data into a single representation, preserving meaningful biological signal in the different modalities and at different resolutions. Moreover, this latent variable space is appropriate for standard downstream analysis techniques commonly used for analyzing cells without any further specialized adjustments, allowing Cobolt to fit into standard analysis pipelines.MethodsWhile the most common application of joint-modality platforms consists of pairs of modalities (such as the example of mRNA-seq and ATAC-seq we described above), we will describe Cobolt in generality, assuming that there are M modalities.Modeling modality dependencyFor an individual cell c, we can (potentially) observe M vectors of data \(x_{c}=\left \{x_{c}^{(1)}, \cdots, x_{c}^{(M)}\right \}\), each vector of dimension d1,…,dM corresponding to the number of features of each modality. We assume a Bayesian latent model, such that for each cell there is a latent variable zc∈RK representing the biological signal of the cell, where zc is assumed drawn from a Gaussian prior distribution. Given zc, we assume that the data \(x_{c}^{(m)}\) for each modality has an independent generative process, potentially different for each modality. Specifically, we assume that the data \(x_{c}^{(m)}\) from each modality are conditionally independent given the common latent variable zc. That is,
$$p\left(x_{c}^{(1)}, \cdots, x_{c}^{(M)}, z_{c}\right) = p(z_{c})\prod_{i=1}^{M} p\left(x_{c}^{(i)}|z_{c}\right). $$ We use \(q\left (z_{c}|x_{c}^{(1)}, \cdots, x_{c}^{(M)}\right)\) as a variational approximation of the posterior distribution \(p\left (z_{c}|x_{c}^{(1)}, \cdots, x_{c}^{(M)}\right)\). \(q\left (z_{c}|x_{c}^{(1)}, \cdots, x_{c}^{(M)}\right)\), the encoder, is assumed Gaussian with parameters modeled as neural networks (i.e., Variational Autoencoder, VAE [41]). This allows for estimation of the posterior distribution \(p\left (z_{c}|x_{c}^{(1)}, \cdots, x_{c}^{(M)}\right)\) and the underlying latent variable for each cell c. The posterior mean of this distribution \(\hat {z}_{c}\) will be our summary of the shared representation across modalities.Importantly, this model can be estimated even when not all of the input data contains all modalities. In this case, an individual cell c contains a subset of the modalities, \(\mathcal {S}_{c}\subset \{1,\ldots,M\}\), and consists of data \(x_{c}=\left \{x_{c}^{(i)},i\in \mathcal {S}_{c}\right \}\). Without all modalities observed, the cell can contribute to the estimation of the model as its distribution can be explicitly written out:
$$p(x_{c}, z_{c}) = p(z_{c})\prod_{i\in \mathcal{S}_{c}} p\left(x_{c}^{(i)}|z_{c}\right). $$ Furthermore, we can estimate latent variables for such cells by using posterior distribution of zc when conditioning only on the observed modalities, \(q\left (z_{c}|\left \{x_{c}^{(i)},i\in \mathcal {S}_{c}\right \}\right)\). Instead of using separate neural networks for 2M−1 posterior distributions of different modality combinations, we adopt a technique introduced in Multimodal Variational Autoencoder (MVAE) [17], which largely reduces the number of encoders to 2M (See Additional file 2: Supplementary Methods for inference details).As an example, if there are two modalities, mRNA-seq and ATAC-seq, and we have n1 cells with paired data from the joint modality platform, \(X_{1}=\left (X_{1}^{mRNA}, X_{1}^{ATAC}\right)\); n2 cells with only mRNA measured, \(X_{2}=X_{2}^{mRNA}\); and n3 cells with only ATAC-seq measured, \(X_{3}=X_{3}^{ATAC}\). All N=n1+n2+n3 cells can be used in the estimation of the joint distribution of the latent variables, and estimates of the latent variables can be found as the mean of the relevant approximate posterior distributions:
$$\begin{array}{*{20}l} \hat{Z}_{1}&=E\left(Z|X_{1}^{mRNA}, X_{1}^{ATAC}\right) &\text{(Paired cells)}\\ \hat{Z}_{2}^{mRNA}&=E\left(Z|X_{2}^{mRNA}\right) &\text{(mRNA-seq only)}\\ \hat{Z}_{3}^{ATAC}&=E\left(Z|X_{3}^{ATAC}\right) &\text{(ATAC-seq only)}\\ \end{array} $$ Correcting for missing modalitiesIn practice, we find that the distributions \(q_{\phi }\left (z_{c}|\left \{x_{c}^{(i)},i\in \mathcal {S}\right \}\right)\) have subtle differences for different subsets \(\mathcal {S}\), i.e., the latent variables \(\hat {Z}_{1}, \hat {Z}_{2}^{mRNA}\), and \(\hat {Z}_{3}^{ATAC}\) show distinct separations (Additional file 1: Fig. S15). One possibility could be due to platform differences between the different datasets that remain even after our batch correction. However, we also see differences in these distributions even if we only consider the joint-modality data, where we can estimate all of these posterior distributions on the same cells, \(\hat {Z}_{1}^{mRNA}=E\left (Z|X_{1}^{mRNA}\right)\) or \(\hat {Z}_{1}^{ATAC}=E\left (Z|X_{1}^{ATAC}\right)\) (Additional file 1: Fig. S15). Indeed, there is nothing in the optimization of the posterior distribution that requires these different posterior distributions to be the same.While the effects are small, these subtle differences can affect downstream analyses, e.g., in clustering cells for subtype discovery. Rather than directly force these posterior distributions to match in our estimation of the model, Cobolt fits the model as described above (using all of the data) and then uses the paired data to train a prediction models that predict \(\hat {Z}_{1}\) from the modality-specific estimates \(\hat {Z}_{1}^{mRNA}\) and \(\hat {Z}_{1}^{ATAC}\). We then apply these prediction models to \(\hat {Z}_{2}^{mRNA}\) and \(\hat {Z}_{3}^{ATAC}\) to obtain estimates \(\hat {Z}_{2}\) and \(\hat {Z}_{3}\) which are better aligned to be jointly analyzed in the same space. In practice, we find XGBoost [42] and k-nearest neighbors algorithm work equally well. We present results based on XGBoost. We would note that there is little difference in performance when we predict coordinates into the ATAC-Seq space E(Z|XATAC) or mRNA-Seq space E(Z|XmRNA), rather than the joint space (\(E\left (Z|X_{1}^{mRNA}, X_{1}^{ATAC}\right)\)), see Additional file 1: Fig. S16.Modeling single modality of sparse countsThe choice of the generative distribution pψ(x(i)|z) should be chosen to reflect the data and in principle can vary from modality to modality. For example, single-modality VAE models using zero-inflated negative binomial distributions (ZINB) [43] have been proposed for scRNA-seq datasets to account for sparse count data. However, we found ZINB models performed less well for technologies that measure modalities like chromatin accessibility, which results in data with sparser counts and larger feature sizes than scRNA-seq. Therefore, we develop a latent model for these types of modalities inspired by the Latent Dirichlet Allocation (LDA) [18].Our generative model for a single modality i starts by assuming that the counts measured on a cell are the mixture of the counts from different latent categories. In the genomic setting, these categories could correspond to biological processes of the cell, for example. Each category has a corresponding distribution of feature counts. The cumulative feature counts for a cell c are then the result of combining the counts across its categories, i.e., a mixture of the categories’ distributions. Specifically, each cell c has a latent probability vector θc∈[0,1]K describing the proportion of each category that makes up cell c. Each category k has a probability vector σ(βk) that provides the distribution of its feature counts. Here, σ indicates the softmax function that transforms βk to a probability vector that sums to 1. The observed vector of counts xc is a multinomial draw with probabilities πc, where πc=σ(B)θc, and B=(β1,…,βK) is a matrix of the individual βk vectors. To extend this model to multiple modalities, we assume a shared latent variable zc that is common across modalities, but each modality has a different B(i) that transforms the shared latent class probabilities into the feature space of the modality.Furthermore, it is well known that there can be meaningful technical artifacts (“batch effects”) between different datasets on the same modality, for example due to differences between platforms or laboratory preparations. To counter this, our model also adjusts the sampling probabilities σ(B(i))θc differently for data from different batches within the same modality i. We would note that the model can also take batch-corrected counts as input, such as are available for mRNA expression data (e.g., [44–46]), but we anticipate that for some modality types stand-alone batch correction techniques may not be as well developed. We evaluate the effect of our batch correction on the 10x Multiome data, which consists of two runs of 10x Multiome on the different patient input sampled at different times. This creates a batch effect between the multi-modality input and the single-modality input where we know the ground truth of how the single-modality data should be linked. We use the same nearest-neighbor analysis as in Fig. 5B, with and without the batch correction terms and see much improved performance using the batch correction (Additional file 1: Fig. S17). This type of quantitative nearest-neighbor analysis is not possible for the SNARE-Seq data, since we do not have two different batches of paired multi-modality data, but we visually see large improvement due to the batch correction when analyzing the single-modality datasets jointly with the multi-modality data (Additional file 1: Fig. S18).The parameter θc is the latent variable describing the contributions of each category to cell c and is shared across all modalities. In LDA models, it is typically assumed to have a Dirichlet prior distribution. However, we use a Laplace approximation to the Dirichlet introduced in ProdLDA [23], which allows for incorporation into a VAE model. This prior assumes a latent variable zc with a Gaussian prior and sets θc=σ(zc), where σ is the softmax transformation. We use this approximation to the Dirichlet distribution to provide a multi-modality method appropriate for sparse sequence count data.Data processing and method implementationSNARE-seq processing and annotationWe downloaded the processed counts of the adult mouse cerebral cortex data (10,309 cells) [8]. We applied quality filtering that retained cells having a number of genes detected greater than 20. For genes, we used the ones detected in more than 5 cells and have a total number of counts greater than 10. For peaks, we removed the ones having nonzero counts in more than 10% of cells or less than 5 cells. We performed clustering analysis using Seurat (version 3.2.2) on the gene expression modality. The data were normalized using SCTransform function with default parameters, followed by principal component analysis (PCA) using the default 3000 variable features. Louvain algorithm was applied on the first 50 PCs with the resolution parameter equals 0.65. Cell type annotations are generated on the resulting 15 clusters using the marker genes [31]. We applied cisTopic (version 0.3.0) on the chromatin accessibility data with default parameters. Model selection was conducted based on log-likelihood using runWrapLDAModels and selectModel functions, and 30 topics are used in the results.For the integration of SNARE-seq with the MOp data using Cobolt, we map the SNARE-seq counts to the peak set called on the MOp scATAC-seq data. Since peaks are typically called in a dataset-specific manner, the ideal integration strategy would be to redo the peak-calling with all datasets combined. However, in Additional file 1: Fig. S10, we show that our simple alternative of mapping data to peaks called on a different dataset from the same system does not result in significant performance loss for Cobolt.MOp data preprocessingWe downloaded the single-nucleus 10x v3 transcriptome dataset (90,266 cells) and the open chromatin dataset (15,731 cells, sample 171206_3C) [31]. For mRNA-seq quality control, we filtered cells that have less than 200 genes detected or have greater than 5% mitochondrial counts. For ATAC-seq, we utilized the TSSEnrichment and blacklist region reads calculation functionalities in Signac. We subsetted cells with the blacklist ratio less than 0.1, the number of unique molecular identifiers (UMIs) greater than 50, and the TSS enrichment score greater than 2 and less than 20. A total of 88,021 and 13,249 were retained for mRNA-seq and ATAC-seq, separately. To make annotation in the two datasets consistent, we merged the layer 2/3 IT and layer 4/5 IT subclusters in ATAC-seq data. For mRNA-seq, we merged Lamp5, Vip, and Sncg into one CGE cluster. When integrating the MOp datasets with the SNARE-seq data, we used only genes detected in both the scRNA-seq and the SNARE-seq datasets.10X PBMC data preprocessingPBMC datasets were downloaded from the 10X website (Multiome Chromium X, Multiome unsorted, scRNA-seq, scATAC-seq). For chromatin accessibility, we mapped the scATAC-seq reads and the Multiome unsorted reads to the peaks called on the Multiome Chromium X data. No quality filtering was applied to any of the four 10X datasets. Clusters with high mitochondrial expression are identified and annotated as low quality clusters, and clusters with the majority of cells identified by DoubletFinder [47] are annotated as doublet clusters in the downstream clustering analysis.Gene activity calculationGene activity matrix for chromatin accessibility is generated by counting the number of reads overlapping genes and their promoters using BEDOPS [48], where a promoter is defined as the region starting from the transcription start site (TSS) to 3000 base pairs upstream of TSS.
Cobolt network architecture and trainingFor each modality i, the encoder takes as input the log-plus-one transformed counts. We use one fully connected layer of size 128, followed by fully connected layers for mean \(\tilde {\mu }^{(i)}\) and log-variance \(\log \tilde {\Sigma }^{(i)}\). We tried networks with one or two hidden layers of varying sizes and found the results pretty stable. The decoders follow our probability model for sparse counts (see also Additional file 2: Supplementary methods, for details) and do not contain neural networks. We set the parameter of the Dirichlet prior to 50 divided by the number of latent variables K. The actual parameters used for the Gaussian prior are calculated using the Laplace approximation (see also Additional file 2, Supplementary methods, for details). For the ELBO objective, we set the weighting terms \(\lambda ^{\mathcal {A}}\) reciprocal to the number of samples available for modality combination \(\mathcal {A}\). We set the hyperparameter weights η for conditional likelihood terms to 1. Adam optimizer is used, and we select a learning rate of 0.005 after tunning. We adopt a KL cost annealing schedule that linearly increases the weight of the KL term γ from 0 to 1 in the first 30 epochs. During training, we use a batch size of 128 and a fixed number of 100 epochs.We note that the softmax transformation from zc to θc is not a one-to-one transformation. Therefore, we scale \(\hat {z}_{c}\) to mean 0 before the downstream correction, followed by clustering and visualization.In correcting for missing modalities, we predict using XGBoost, setting the objective function to regression with squared loss, the learning rate to 0.8, and the maximum depth of a tree to 3. XGBoost is applied separately to each modality (scRNA-Seq and scATAC-Seq), resulting in our corrected estimates \(\hat {Z}_{2}, \hat {Z}_{3}\).The number of latent variables K was set to 10 in the SNARE-seq analysis and the method comparison analysis so as to be consistent with the default of several other methods for comparison purposes. We used K=30 for the mouse cortex integration and K=10 for the 10X PBMC integration. Estimations of the data’s marginal likelihood were used to assist the selection of K.For the mouse cortex data integration, we focused on genes and peaks that have top 30% average expression and removed the ones in the top 1%. For the SNARE-seq analysis and the 10X PBMC integration, all features were included in training. The choice of top features is less important here, which we found to have a small effect on the results.Clustering and visualization of Cobolt resultsClusters were generated on the corrected latent variables using Louvain algorithm [49]. We used the implementation of naive Louvain algorithm in FindClusters function from the R package Seurat. All parameters, other than the resolution controlling the number of clusters identified, were set to default. The results of the clustering of the integrative analysis of the SNARE-Seq with the MOP scRNA-Seq and ATAC-Seq from the mouse cortex are given in Additional file 3: Table S1 (resolution = 0.8). The results of the clustering of the integrative analysis of the 10X PBMC data are given in Additional file 4: Table S2 (resolution = 0.8). UMAPs were generated using the umap function from the R package uwot with the number of neighbors set to 30.MOFA2, scMM, and BABEL analysis on the SNARE-seq dataFollowing the vignette of MOFA2, we used the top 2500 variable genes and cisTopic embeddings of the chromatin accessibility modality as input. Variable genes are selected using the FindVariableFeatures function from Seurat using selection method “vst”. The number of factors was set to 10. Two factors were identified as technical factors after inspecting their correlation with the total number of reads counts per cell. The UMAP representation was then generated using the rest of the factors with the number of neighbors set to 30. scMM was run with batch size equals 32, number of epochs equals 50, learning rate equals 10−4, number of latent dimensions equals 10, number of hidden dimension for gene expression equals 100, and number of hidden dimensions for chromatin accessibility equals 500. The parameters were chosen following the scMM paper. BABEL was run with the number of latent dimensions set to 10 and the batch size set to 256. Other parameters were chosen to follow the built-in SNARE-seq defaults.Differential analysisDE analysis on gene expression and DA analysis on gene activities were performed using the Wilcoxon rank sum test followed by Bonferroni correction, implemented by the FindMarkers function in Seurat. Gene expression and gene activities were visualized by heatmaps using DoHeatmap in Seurat. DA analysis on the peaks for the SNARE-seq analysis was performed using Fisher’s exact test followed by the Benjamini–Hochberg procedure (following [8]). Peak-by-cluster matrix was normalized by size factors calculated by Monocle [50] and visualized by heatmaps (following [8]). Genes and peaks with adjusted p-values lower than 0.05 were called significant.LIGER and Signac analysis on the MOp dataLIGER and Signac (Seurat) take as input gene-level count summaries from different modalities, such as gene expression or gene body methylation/chromatin accessibility measures. The input is different from Cobolt, which uses the peak summaries directly without needing to summarize at the gene level. Therefore, we applied these two methods on the gene expression and gene activity matrices, where the latter is defined as the summarized chromatin accessibility counts over gene and promoter regions.We ran LIGER (version 0.5.0) using default parameters on the filtered data. Parameter K for factorization is set to 30 after inspecting the plot generated by function suggestK. Louvain clustering was performed by setting the resolution such that 17 clusters were obtained. We ran Signac (version 1.1.0) on the same filtered data. We first performed clustering analysis on the gene expression modality and then transferred the cluster labels to the open chromatin modality. Both the gene expression matrix and gene activity matrix were normalized by running NormalizeData followed by ScaleData. For the gene activity matrix, the scale.factor parameter was set to the median of UMI distribution as suggested by the vignette. Other parameters were kept as default. We then ran FindTransferAnchors with the default number of dimensions equals 30, which is the same as used for LIGER. Finally, we ran TransferData with weight reduction set to “cca” or the Latent Semantic Indexing (LSI) from analyzing the peak matrix. Results using LSI are presented in the paper as it performed better than the “cca” option.Test training split for method comparisonFor scMM and BABEL, which do not allow the single-modality data to be used in the training of the model, we assign 20% of the cells as paired modality data that is used for the training set; the trained model was then used to generate the embedding on the rest of the cells without providing the pairing information. This provided separate estimates for the unpaired mRNA and ATAC modalities, respectively, with which we evaluated whether the paired cells were close together. For Cobolt, which allows the use of single-modality data for the training of the model, we assign 20% of the cells as paired modality data and the other 80% were given as single-modality data, and then we trained the model on all of the cells. For LIGER and Signac, which are not designed for multiple-modality data, we hide the pairing information on all cells and treat all of the cells as if they were collected on different cells. To be comparable with the other methods, we only evaluated their performance on the 80% of cells treated by the other methods as single-modality data.
Availability of data and materials
An open-source package Cobolt, implemented in Python, is available on GitHub (https://github.com/boyinggong/cobolt) under the GNU General Public License v3.0; the analysis for this manuscript was done with v1.0.0 [51]. Code scripts for reproducing results presented in the paper are publicly accessible on GitHub (https://github.com/boyinggong/cobolt_manuscript) under the GNU General Public License v3.0 [52].
Datasets used in this paper are all publicly available. The SNARE-seq data can be downloaded from Gene Expression Omnibus with accession number GSE126074. The MOp scRNA-seq and scATAC-seq data can be downloaded from NeMO Archive with accession number nemo:dat-ch1nqb7. The 10X datasets are available on the 10x Genomics website (Multiome Chromium X, Multiome unsorted, scRNA-seq, scATAC-seq).
ReferencesWelch JD, Kozareva V, Ferreira A, Vanderburg C, Martin C, Macosko EZ. Single-cell multi-omic integration compares and contrasts features of brain cell identity. Cell. 2019; 177(7):1873–87.Article
PubMed
PubMed Central
CAS
Google Scholar
Stuart T, Srivastava A, Madad S, et al.Single-cell chromatin state analysis with Signac. Nat Methods. 2021; 18:1333–41. https://doi.org/10.1038/s41592-021-01282-5.Article
PubMed
CAS
Google Scholar
Stuart T, Butler A, Hoffman P, Hafemeister C, Papalexi E, Mauck III WM, Hao Y, Stoeckius M, Smibert P, Satija R. Comprehensive integration of single-cell data. Cell. 2019; 177(7):1888–902.Article
PubMed
PubMed Central
CAS
Google Scholar
Stoeckius M, Hafemeister C, Stephenson W, Houck-Loomis B, Chattopadhyay PK, Swerdlow H, Satija R, Smibert P. Simultaneous epitope and transcriptome measurement in single cells. Nat Methods. 2017; 14(9):865–8.Article
PubMed
PubMed Central
CAS
Google Scholar
Clark SJ, Argelaguet R, Kapourani C-A, Stubbs TM, Lee HJ, Alda-Catalinas C, Krueger F, Sanguinetti G, Kelsey G, Marioni JC, et al.scNMT-seq enables joint profiling of chromatin accessibility DNA methylation and transcription in single cells. Nat Commun. 2018; 9(1):1–9.Article
Google Scholar
Cao J, Cusanovich DA, Ramani V, Aghamirzaie D, Pliner HA, Hill AJ, Daza RM, McFaline-Figueroa JL, Packer JS, Christiansen L, et al.Joint profiling of chromatin accessibility and gene expression in thousands of single cells. Science. 2018; 361(6409):1380–5.Article
PubMed
PubMed Central
CAS
Google Scholar
Zhu C, Yu M, Huang H, Juric I, Abnousi A, Hu R, Lucero J, Behrens MM, Hu M, Ren B. An ultra high-throughput method for single-cell joint analysis of open chromatin and transcriptome. Nat Struct Mol Biol. 2019; 26(11):1063–70.Article
PubMed
PubMed Central
CAS
Google Scholar
Chen S, Lake BB, Zhang K. High-throughput sequencing of the transcriptome and chromatin accessibility in the same cell. Nat Biotechnol. 2019; 37(12):1452–7.Article
PubMed
PubMed Central
CAS
Google Scholar
Lee J, Hwang D, et al.Single-cell multiomics: technologies and data analysis methods. Exp Mol Med. 2020; 52(9):1428–42.Article
PubMed
PubMed Central
CAS
Google Scholar
Gayoso A, Steier Z, Lopez R, et al.Joint probabilistic modeling of single-cell multi-omic data with totalVI. Nat Methods. 2021; 18:272–82. https://doi.org/10.1038/s41592-020-01050-x.Article
PubMed
PubMed Central
CAS
Google Scholar
Wang X, Sun Z, Zhang Y, Xu Z, Xin H, Huang H, Duerr RH, Chen K, Ding Y, Chen W. BREM-SC: a Bayesian random effects mixture model for joint clustering single cell multi-omics data. Nucleic Acids Res. 2020; 48(11):5814–24.Article
PubMed
PubMed Central
CAS
Google Scholar
Hao Y, Hao S, Andersen-Nissen E, Mauck WM, Zheng S, Butler A, Lee MJ, Wilk AJ, Darby C, Zager M, Hoffman P, Stoeckius M, Papalexi E, Mimitou EP, Jain J, Srivastava A, Stuart T, Fleming LM, Yeung B, Rogers AJ, McElrath JM, Blish CA, Gottardo R, Smibert P, Satija R. Integrated analysis of multimodal single-cell data. Cell. 2021; 184(13):3573–3587.e29. https://doi.org/10.1016/j.cell.2021.04.048.Article
PubMed
PubMed Central
CAS
Google Scholar
Argelaguet R, Arnol D, Bredikhin D, Deloro Y, Velten B, Marioni JC, Stegle O. MOFA+: a statistical framework for comprehensive integration of multi-modal single-cell data. Genome Biol. 2020; 21:1–17.Article
Google Scholar
Zuo C, Chen L. Deep-joint-learning analysis model of single cell transcriptome and open chromatin accessibility data. Brief Bioinforma. 2021; 22(4):287.Article
Google Scholar
Wu KE, Yost KE, Chang HY, Zou J. BABEL enables cross-modality translation between multiomic profiles at single-cell resolution. Proc Natl Acad Sci. 2021; 118(15):e2023070118. https://doi.org/10.1073/pnas.2023070118.Article
PubMed
PubMed Central
CAS
Google Scholar
Minoura K, Abe K, Nam H, Nishikawa H, Shimamura T. A mixture-of-experts deep generative model for integrated analysis of single-cell multiomics data. Cell Rep Methods. 2021; 1(5):100071. https://doi.org/10.1016/j.crmeth.2021.100071.Article
PubMed Central
Google Scholar
Wu M, Goodman N. Multimodal Generative Models for Scalable Weakly-Supervised Learning In: Bengio S, Wallach H, Larochelle H, Grauman K, Cesa-Bianchi N, Garnett R, editors. Advances in Neural Information Processing Systems. Curran Associates, Inc.: 2018. https://proceedings.neurips.cc/paper/2018/file/1102a326d5f7c9e04fc3c89d0ede88c9-Paper.pdf .Blei DM, Ng AY, Jordan MI. Latent dirichlet allocation. J Mach Learn Res. 2003; 3(Jan):993–1022.
Google Scholar
Yotsukura S, Nomura S, Aburatani H, Tsuda K, et al.CellTree: an R/bioconductor package to infer the hierarchical structure of cell populations from single-cell RNA-seq data. BMC Bioinforma. 2016; 17(1):363.Article
Google Scholar
Sun Z, Wang T, Deng K, Wang X-F, Lafyatis R, Ding Y, Hu M, Chen W. DIMM-SC: a Dirichlet mixture model for clustering droplet-based single cell transcriptomic data. Bioinformatics. 2018; 34(1):139–46.Article
PubMed
CAS
Google Scholar
González-Blas CB, Minnoye L, Papasokrati D, Aibar S, Hulselmans G, Christiaens V, Davie K, Wouters J, Aerts S. cisTopic: cis-regulatory topic modeling on single-cell ATAC-seq data. Nature Methods. 2019; 16(5):397–400.Article
PubMed Central
Google Scholar
Backenroth D, He Z, Kiryluk K, Boeva V, Pethukova L, Khurana E, Christiano A, Buxbaum JD, Ionita-Laza I. FUN-LDA: a latent Dirichlet allocation model for predicting tissue-specific functional effects of noncoding variation: methods and applications. Am J Hum Genet. 2018; 102(5):920–42.Article
PubMed
PubMed Central
CAS
Google Scholar
Srivastava A, Sutton C. Autoencoding Variational Inference for Topic Models. In: Proceedings for the 5th International Conference on Learning Representations: 2017. https://iclr.cc/archive/www/2017.html. 5th International Conference on Learning Representations, ICLR 2017 ; Conference date: 24-04-2017 Through 26-04-2017.Tasic B, Menon V, Nguyen TN, Kim TK, Jarsky T, Yao Z, Levi B, Gray LT, Sorensen SA, Dolbeare T, et al.Adult mouse cortical cell taxonomy revealed by single cell transcriptomics. Nat Neurosci. 2016; 19(2):335–46.Article
PubMed
PubMed Central
CAS
Google Scholar
McInnes L, Healy J, Melville J. Umap: uniform manifold approximation and projection for dimension reduction. arXiv preprint arXiv:1802.03426. 2018. https://github.com/lmcinnes/umap.Baker A, Kalmbach B, Morishima M, Kim J, Juavinett A, Li N, Dembrow N. Specialized subpopulations of deep-layer pyramidal neurons in the neocortex: bridging cellular properties to functional consequences. J Neurosci. 2018; 38(24):5441–55. https://doi.org/10.1523/JNEUROSCI.0150-18.2018.Article
PubMed
PubMed Central
CAS
Google Scholar
Sorensen SA, Bernard A, Menon V, Royall JJ, Glattfelder KJ, Desta T, Hirokawa K, Mortrud M, Miller JA, Zeng H, Hohmann JG, Jones AR, Lein ES. Correlated gene expression and target specificity demonstrate excitatory projection neuron diversity. Cereb Cortex. 2015; 25(2):433–449. https://doi.org/10.1093/cercor/bht243.Article
PubMed
Google Scholar
Belgard TG, Marques AC, Oliver PL, Abaan HO, Sirey TM, Hoerder-Suabedissen A, García-Moreno F, Molnár Z, Margulies EH, Ponting CP. A transcriptomic atlas of mouse neocortical layers. Neuron. 2011; 71(4):605–16. https://doi.org/10.1016/j.neuron.2011.06.039.Article
PubMed
PubMed Central
CAS
Google Scholar
Zeng H, Shen EH, Hohmann JG, Oh SW, Bernard A, Royall JJ, Glattfelder KJ, Sunkin SM, Morris JA, Guillozet-Bongaarts AL, Smith KA, Ebbert AJ, Swanson B, Kuan L, Page DT, Overly CC, Lein ES, Hawrylycz MJ, Hof PR, Hyde TM, Kleinman JE, Jones AR. Large-scale cellular-resolution gene profiling in human neocortex reveals species-specific molecular signatures. Cell. 2012; 149(2):483–96. https://doi.org/10.1016/j.cell.2012.02.052.Article
PubMed
PubMed Central
CAS
Google Scholar
Fazel Darbandi S, Robinson Schwartz SE, Qi Q, Catta-Preta R, Pai ELL, Mandell JD, Everitt A, Rubin A, Krasnoff RA, Katzman S, Tastad D, Nord AS, Willsey AJ, Chen B, State MW, Sohal VS, Rubenstein JLR. Neonatal Tbr1 dosage controls cortical layer 6 connectivity. Neuron. 2018; 100(4):831–8457. https://doi.org/10.1016/j.neuron.2018.09.027.Article
PubMed
CAS
Google Scholar
Yao Z, Liu H, Xie F, Fischer S, Adkins RS, Aldridge AI, Ament SA, Bartlett A, Behrens MM, Van den Berge K, Bertagnolli D, de Bézieux HR, Biancalani T, Booeshaghi AS, Bravo HC, Casper T, Colantuoni C, Crabtree J, Creasy H, Crichton K, Crow M, Dee N, Dougherty EL, Doyle WI, Dudoit S, Fang R, Felix V, Fong O, Giglio M, Goldy J, Hawrylycz M, Herb BR, Hertzano R, Hou X, Hu Q, Kancherla J, Kroll M, Lathia K, Li YE, Lucero JD, Luo C, Mahurkar A, McMillen D, Nadaf NM, Nery JR, Nguyen TN, Niu S-Y, Ntranos V, Orvis J, Osteen JK, Pham T, Pinto-Duarte A, Poirion O, Preissl S, Purdom E, Rimorin C, Risso D, Rivkin AC, Smith K, Street K, Sulc J, Svensson V, Tieu M, Torkelson A, Tung H, Vaishnav ED, Vanderburg CR, van Velthoven C, Wang X, White OR, Huang ZJ, Kharchenko PV, Pachter L, Ngai J, Regev A, Tasic B, Welch JD, Gillis J, Macosko EZ, Ren B, Ecker JR, Zeng H, Mukamel EA. A transcriptomic and epigenomic cell atlas of the mouse primary motor cortex. Nature. 2021; 598(7879):103–110. https://doi.org/10.1038/s41586-021-03500-8.Article
PubMed
PubMed Central
CAS
Google Scholar
10x Genomics. Pbmcs from human (multiome v1.0, chromium x), single cell multiome atac + gene expression dataset by cell ranger arc 2.0.0. 2021.10x Genomics. Pbmcs from human (no cell sorting, chromium next gem), single cell multiome atac + gene expression dataset by cell ranger arc 2.0.0. 2021.10x Genomics. Pbmcs from human (3’ ht v3.1, chromium x), single cell gene expression dataset by cell ranger 6.1.0. 2021.10x Genomics. Pbmcs from human (atac v1.1, chromium x), single cell atac dataset by cell ranger atac 2.0.0. 2021.Ding J, Adiconis X, Simmons SK, Kowalczyk MS, Hession CC, Marjanovic ND, Hughes TK, Wadsworth MH, Burks T, Nguyen LT, et al.Systematic comparison of single-cell and single-nucleus RNA-sequencing methods. Nat Biotechnol. 2020; 38(6):737–746.Article
PubMed
PubMed Central
CAS
Google Scholar
Pliner HA, Shendure J, Trapnell C. Supervised classification enables rapid annotation of cell atlases. Nat Methods. 2019; 16(10):983–6.Article
PubMed
PubMed Central
CAS
Google Scholar
Franzén O, Gan L-M, Björkegren JLM. PanglaoDB: a web server for exploration of mouse and human single-cell RNA sequencing data. Database. 2019; 2019:baz046. https://doi.org/10.1093/database/baz046.Article
PubMed
PubMed Central
Google Scholar
Tasic B, Yao Z, Graybuck LT, Smith KA, Nguyen TN, Bertagnolli D, Goldy J, Garren E, Economo MN, Viswanathan S, Penn O, Bakken T, Menon V, Miller J, Fong O, Hirokawa KE, Lathia K, Rimorin C, Tieu M, Larsen R, Casper T, Barkan E, Kroll M, Parry S, Shapovalova NV, Hirschstein D, Pendergraft J, Sullivan HA, Kim TK, Szafer A, Dee N, Groblewski P, Wickersham I, Cetin A, Harris JA, Levi BP, Sunkin SM, Madisen L, Daigle TL, Looger L, Bernard A, Phillips J, Lein E, Hawrylycz M, Svoboda K, Jones AR, Koch C, Zeng H. Shared and distinct transcriptomic cell types across neocortical areas. Nature. 2018; 563(7729):72–8. https://doi.org/10.1038/s41586-018-0654-5.Article
PubMed
PubMed Central
CAS
Google Scholar
Saelens W, Cannoodt R, Todorov H, Saeys Y. A comparison of single-cell trajectory inference methods. Nat Biotechnol. 2019; 37(5):547–54. https://doi.org/10.1038/s41587-019-0071-9.Article
CAS
PubMed
Google Scholar
Kingma DP, Welling M. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114. 2013.Chen T, Guestrin C. XGBoost: A Scalable Tree Boosting System. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD ’16). New York: Association for Computing Machinery: 2016. p. 785–94. https://doi.org/10.1145/2939672.2939785.
Google Scholar
Lopez R, Regier J, Cole MB, Jordan MI, Yosef N. Deep generative modeling for single-cell transcriptomics. Nat Methods. 2018; 15(12):1053–8.Article
PubMed
PubMed Central
CAS
Google Scholar
Risso D, Schwartz K, Sherlock G, Dudoit S. GC-content normalization for RNA-Seq data. BMC Bioinforma. 2011; 12(1):480. https://doi.org/10.1186/1471-2105-12-480.Article
CAS
Google Scholar
Zhang Y, Parmigiani G, Johnson WE. ComBat-seq: batch effect adjustment for RNA-seq count data. NAR Genom Bioinforma. 2020; 2(3). https://doi.org/10.1093/nargab/lqaa078.Cole MB, Risso D, Wagner A, DeTomaso D, Ngai J, Purdom E, Dudoit S, Yosef N. Performance assessment and selection of normalization procedures for single-cell RNA-Seq. Cell Syst. 2019; 8(4):315–3288. https://doi.org/10.1016/j.cels.2019.03.010.Article
PubMed
PubMed Central
CAS
Google Scholar
McGinnis CS, Murrow LM, Gartner ZJ. DoubletFinder: doublet detection in single-cell RNA sequencing data using artificial nearest neighbors. Cell Syst. 2019; 8(4):329–37.Article
PubMed
PubMed Central
CAS
Google Scholar
Neph S, Kuehn MS, Reynolds AP, Haugen E, Thurman RE, Johnson AK, Rynes E, Maurano MT, Vierstra J, Thomas S, et al.BEDOPS: high-performance genomic feature operations. Bioinformatics. 2012; 28(14):1919–20.Article
PubMed
PubMed Central
CAS
Google Scholar
Blondel VD, Guillaume J-L, Lambiotte R, Lefebvre E. Fast unfolding of communities in large networks. J Stat Mech Theory Exp. 2008; 2008(10):10008.Article
Google Scholar
Qiu X, Hill A, Packer J, Lin D, Ma Y-A, Trapnell C. Single-cell mRNA quantification and differential analysis with Census. Nat Methods. 2017; 14(3):309–15.Article
PubMed
PubMed Central
CAS
Google Scholar
Gong B, Purdom E. cobolt. Github. 2021. https://doi.org/10.5281/zenodo.5714790.Gong B. cobolt_manuscript. Github. 2021. https://doi.org/10.5281/zenodo.5715087.Download referencesAcknowledgementsThe authors would like to thank Jean-Philippe Vert for useful early discussions on the analysis of multi-modality data and Hector Roux de Bezieux for assistance in accessing the MOp data.
Peer review information
Barbara Cheifet was the primary editor of this article and managed its editorial process and peer review in collaboration with the rest of the editorial team.
Review history
The review history is available as Additional file 5.
FundingThis work has been supported by the NIH grant U19MH114830.Author informationAuthors and AffiliationsDivision of Biostatistics, University of California, Berkeley, Berkeley, CA, USABoying Gong & Yun ZhouDepartment of Statistics, University of California, Berkeley, Berkeley, CA, USAElizabeth PurdomAuthorsBoying GongView author publicationsYou can also search for this author in
PubMed Google ScholarYun ZhouView author publicationsYou can also search for this author in
PubMed Google ScholarElizabeth PurdomView author publicationsYou can also search for this author in
PubMed Google ScholarContributionsEP formulated the problem and supervised the work. BG developed the model and performed the analysis, with inputs from all authors. BG and YZ developed the model inference and coded the initial implementation. BG implemented the package. BG and EP drafted and revised the manuscript. YZ revised the manuscript and figures. All authors read and approved the final manuscript.Corresponding authorCorrespondence to
Elizabeth Purdom.Ethics declarations
Ethics approval and consent to participate
Not applicable.
Consent for publication
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Additional informationPublisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Supplementary InformationAdditional file 1 Supplementary Figures. PDF file containing Supplementary Figures referenced in the main text.Additional file 2 Supplementary Methods. PDF file containing additional details regarding the Cobolt algorithm.Additional file 3 Table S1. CSV file containing the cell clustering of the mouse cortex data integration.Additional file 4 Table S2. CSV file containing the cell clustering of the 10X PBMC integration.Additional file 5 Review history.Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Reprints and permissionsAbout this articleCite this articleGong, B., Zhou, Y. & Purdom, E. Cobolt: integrative analysis of multimodal single-cell sequencing data.
Genome Biol 22, 351 (2021). https://doi.org/10.1186/s13059-021-02556-zDownload citationReceived: 15 July 2021Accepted: 22 November 2021Published: 28 December 2021DOI: https://doi.org/10.1186/s13059-021-02556-zShare this articleAnyone you share the following link with will be able to read this content:Get shareable linkSorry, a shareable link is not currently available for this article.Copy to clipboard
Provided by the Springer Nature SharedIt content-sharing initiative
KeywordsSingle cellMulti-omicsIntegration
Download PDF
Advertisement
Genome Biology
ISSN: 1474-760X
Contact us
Submission enquiries: editorial@genomebiology.com
General enquiries: info@biomedcentral.com
Read more on our blogs
Receive BMC newsletters
Manage article alerts
Language editing for authors
Scientific editing for authors
Policies
Accessibility
Press center
Support and Contact
Leave feedback
Careers
Follow BMC
BMC Twitter page
BMC Facebook page
BMC Weibo page
By using this website, you agree to our
Terms and Conditions,
Your US state privacy rights,
Privacy
statement and
Cookies policy.
Your privacy choices/Manage cookies we use in the preference centre.
© 2024 BioMed Central Ltd unless otherwise stated. Part of
Springer Nature.
Cobolt-上海星谱科技有限公司
Cobolt-上海星谱科技有限公司
首页
产品中心
产品分类
新闻
联系我们
加入我们
探索光的世界…… 021-8010 2555
首页
产品中心
产品分类
新闻
联系我们
加入我们
Cobolt
首页
Cobolt
Cobolt Rogue系列单横模大功率激光器
高功率DPSS连续激光器
波长：640nm
功率：1W
应用：超分辨显微
了解详情
Cobolt多波长激光方案
星谱科技可提供各种配置的激光引擎，输出多个波长的激光用于拉曼，全息，光遗传，TERS等领域。
了解详情
Cobolt 04-01系列单频激光器
单纵模连续DPSS激光器
波长：457-1064 nm
功率：25-100 mW
应用：拉曼，荧光显微，LDV，DLS
了解详情
Cobolt 05-01系列单频激光器
高功率、单纵模DPSS连续激光器
波长：320-1064 nm
功率：10 mW-3 W
应用：全息，拉曼，荧光显微，流式细胞仪
了解详情
Cobolt 06-01半导体激光器
即插即用、可调制连续激光器
波长：375-1064 nm
功率：40-400 W
应用：荧光显微，流式细胞仪，光遗传
了解详情
Cobolt Skyra多波长激光器
革命性多波长激光平台
波长：405-685 nm
功率：50mW, 100mW
应用：荧光显微，流式细胞仪
了解详情
Cobolt 08-01窄线宽激光器
紧凑型、窄线宽连续激光器
波长：405 -1064 nm
功率：25 - 500 mW
应用：拉曼，激光干涉，动态光散射
了解详情
Cobolt 中红外OPO ns激光器
中红外可调谐脉冲激光器
波长：3264-4330 nm
功率：60-80 mW, 10kHz
应用：PAS, LIBS, 光声成像
了解详情
Cobolt 调Q纳秒激光器
高性能调Q脉冲激光器
波长：355, 532, 1064 nm
能量：50-150 μj，重频至7kHz
应用：LIBS, Maldi-TOF, 微加工，打标
了解详情
HÜBNER C-FLEX激光合束器
灵活、紧凑型激光合束器
波长：375-1064 nm
功率：50-1000 mW
应用：共聚焦显微，拉曼，全息
了解详情
Product Categories Acton (24)
Alluxa (1)
Andor (11)
Andor低温恒温器 (0)
Andor光谱仪 (1)
Andor分屏器 (3)
Andor显微镜 (2)
Andor相机 (4)
Argolight (8)
AsahiFilter (5)
AXIS (9)
AZURLIGHT (3)
Becker&Hickl GmbH (14)
C-WAVE (2)
Chroma (3)
Chroma-偏振片 (1)
Chroma-分束镜 (1)
Clean-optical-surface (1)
Cobolt (10)
Cobolt 04 (1)
Cobolt 05 (1)
Cobolt 06 (1)
Cobolt 08 (1)
Cobolt C-FLEX (1)
Cobolt Odin (1)
Cobolt Rogue (1)
Cobolt Skyra (1)
Cobolt Tor (1)
Energetiq (3)
Fergie (1)
FLIR相机 (1)
Hubner (0)
IDQ (7)
Clavis3 QKD Platform (0)
ID Qube (1)
ID100 (1)
ID1000 (1)
ID101 (0)
ID120 (1)
ID230 (1)
ID281 (1)
ID3000 (1)
Inprentus (1)
liquidinstruments (1)
Lumencor (11)
Monstr sense (0)
NLIR (5)
2-5μm红外光谱仪 (1)
7.6-12μm红外光谱仪 (1)
单波长红外探测器 (1)
波长转换模块 (1)
热红外光源 (1)
NOST (1)
Pi imaging (1)
Quantum Light Instruments (3)
Refined (1)
Semrock (0)
SLI (9)
光波导 (0)
光纤光谱仪 (1)
光路调节尺 (1)
光路调节尺 (0)
卤素灯光源 (1)
波长可调光源 (3)
波长选择器 (1)
高光谱成像 (2)
SLM (1)
SmarAct (16)
Sparrow Quantum (0)
Spectrogon (9)
Superlum (3)
TCSPC (5)
BH-TCSPC (5)
Toptica (0)
VALO (1)
WITec (0)
ZAURLIGHT (0)
产品引用 (0)
拉曼 (0)
低温恒温器 (2)
信号处理器件 (1)
IMARIS (0)
前置放大器 (0)
反相器 (0)
衰减器 (0)
锁相放大器 (1)
光学元器件 (43)
SERS芯片 (1)
光栅 (2)
光路调节尺 (0)
显微镜标准片 (8)
波长选择器 (3)
滤光片 (42)
像素滤光片 (0)
其他镜片类 (0)
可调滤光片 (1)
多光谱滤光片 (0)
拉曼滤光片 (0)
窄带滤光片 (1)
紫外光学基材 (4)
紫外反射镜和窗口 (5)
紫外滤光片 (7)
紫外激光选择 (7)
紫外相机响应拓展 (1)
红外滤光片 (8)
荧光滤光片 (1)
衰减片 (4)
光学系统能力 (0)
FLIM/PLIM信号采集软件 (0)
光学系统设计 (0)
图像处理软件 (0)
成像光学设计 (0)
激光雷达信号处理软件 (0)
非成像光学设计 (0)
光机械及精密计量传感 (14)
微纳位移控制 (13)
微纳位移控制及光机 (16)
皮米干涉仪 (0)
纳米定位器 (14)
SmarAct控制器 (1)
微夹钳 (1)
旋转位移台 (1)
电动光阑 (1)
电动滤光片轮 (1)
电动镜架 (1)
线性位移台 (8)
角位移台 (1)
光源 (43)
多波长光源 (14)
C-FLEX 4-8波长 (2)
Cobolt Skyra 4波长 (2)
Cobolt双波长 (1)
LED光源 (0)
Lumencor 多色激光 (2)
Lumencor固态多色 (5)
宽谱光源 (15)
LDLS光源 (2)
LED光源 (1)
SLD（SLED） (2)
卤钨灯 (1)
固态多色光源 (6)
固态白光源 (3)
超连续谱激光 (2)
波长可调激光器及光源 (10)
C-WAVE 单频OPO激光器 (2)
EQ CSE光源 (1)
EQ LDTLS光源 (1)
Superlum 波长可调谐半导体激光器 (1)
波长可调光源 (2)
超连续谱激光 、波长选择器 (2)
激光器 (28)
Refined Laser (1)
中红外脉冲OPO激光器 (1)
半导体激光器 (9)
单频OPO激光器 (2)
单频激光器 (10)
多波长ps激光器 (1)
多波长连续激光器 (5)
波长可调谐半导体激光器 (1)
皮秒激光器 (4)
调Q 纳秒激光器 (5)
飞秒激光器 (1)
高功率单频激光器 (3)
脉冲光源、单光子源 (2)
fs激光器 (1)
ns激光器 (5)
ps激光器 (5)
光谱系统 (6)
光纤光谱仪 (1)
多维度光谱平台 (4)
拉曼光谱 (2)
拉曼积分球 (1)
拉曼系统 (3)
时间分辨光谱 (2)
海洋光谱仪 (2)
瞬态吸收光谱 (2)
荧光光谱 (3)
单光子探测器 (12)
23通道SPAD (2)
HPM (1)
PMT (1)
SNSPD (1)
SPAD (8)
单光子相机 (3)
单色仪和光谱仪 (1)
博维光谱仪 (3)
埃赛力达 (1)
成像系统 (7)
FLIM/PLIM成像显微镜 (1)
SRS/CARS&双光子成像显微镜 (2)
拉曼成像显微镜 (0)
时间分辨成像显微镜 (0)
红外显微镜 (1)
荧光成像显微镜 (0)
荧光成像系统 (0)
转盘共聚焦成像显微镜 (2)
高光谱成像显微镜 (0)
拉曼应用 (4)
探测器 (18)
EMCCD (3)
FLIR工业相机 (1)
ICCD (3)
MIR相机 (0)
sCOMS (7)
SWIR相机 (1)
VIS-NIR相机 (0)
Xray相机 (3)
光谱型相机 (2)
分幅相机 (0)
条纹相机 (5)
科学级CCD (1)
紫外相机响应拓展 (1)
超快相机 (3)
推广页 (33)
无掩模光刻机 (1)
时间控制平台 (5)
SDK平台 (0)
TCSPC卡 (1)
时间控制器 (4)
海洋光学 (3)
SERS芯片 (1)
光纤光谱仪 (0)
光谱系统 (2)
显微光谱 (1)
显微拉曼光谱 (2)
激光安全防护 (0)
主动式激光防护系统 (0)
光学平台激光挡光板 (0)
柔性激光防护屏 (0)
激光安全眼镜 (0)
激光安全联锁系统 (0)
激光防护帘 (0)
激光防护房 (0)
激光防护服 (0)
激光防护眼镜 (0)
激光防护窗 (0)
激光防护罩 (0)
激光防护面罩 (0)
硬质激光防护屏 (0)
首页产品 (7)
联系我们上海市普陀区中江路338弄国盛中心2号楼1506室 021-8010 2555/2599159-2135-3011star-dong@dyna-sense.com
公司产品介绍 Semrock：高性能荧光滤光片， 拉曼滤光片，激光反射镜，窄带滤光片
Cobolt：单纵模、窄线宽、高功率DPSS激光器，多波长激光器
Andor Technology：科学级EMCCD，sCMOS相机及共聚焦显微系统
HÜBNER Photonics：OPO激光器，激光合束器，太赫兹成像产品
Becker & Hickl：TCSPC单光子计数器，荧光寿命成像-FLIM系统
id Quantique：TCSPC单光子计数器，SPAD，近红外InGaAs SPAD，超导纳米线探测器，量子传感
SmartAct：微米、纳米移动控制系统，机械手，真空、低温系统用移动台
联系我们上海市普陀区中江路338弄国盛中心2号楼1506室021-8010 2555/2599159-2135-3011021-8010 2599info@star-spectrum.comCopyright © 上海星谱科技有限公司沪ICP备2021003575号-1
首页
COMPANY
关于我们
产品
新闻
新闻
联系我们
Text Widget Lorem ipsum dolor sit amet, consectetur adipiscing elit. Vivamus quis neque vel quam fringilla feugiat. Suspendisse potenti. Proin eget ex nibh. Nullam convallis tristique pellentesque.