Recommended
More Related Content
What's hot
What's hot (20)
Similar to Integrated genetic and transcriptional analysis at the single-cell level
Similar to Integrated genetic and transcriptional analysis at the single-cell level (20)
Recently uploaded
Recently uploaded (20)
Integrated genetic and transcriptional analysis at the single-cell level
- 1. Jean Fan | ISMB 2018 1 Integrated genetic and transcriptional analysis at the single-cell level Jean Fan NCI F99/K00 Fellow Zhuang Lab Harvard University
- 2. Jean Fan | ISMB 2018 2 Introduction
- 3. Heterogeneity exists within individual patients (intra-tumoral heterogeneity) Jean Fan | ISMB 2018 3 Valent P, Bonnet D, De maria R, et al. Cancer stem cell definitions and terminology: the devil is in the details. Nat Rev Cancer. 2012;12(11):767-75.
- 4. Intra-tumoral heterogeneity presents challenges to current treatment standards 4 Subclonal proportions change over time! Indicative of an active evolutionary process! Jean Fan | ISMB 2018
- 5. Intra-tumoral heterogeneity presents challenges to current treatment standards 5 Subclonal frequencies change over time! Indicative of an active evolutionary process! Jean Fan | ISMB 2018
- 6. Single cell measurements and analysis is needed to understand varying levels of heterogeneity 6 Scenario 1 (co-occurring) Scenario 2 (mutually exclusive) or Jean Fan | ISMB 2018
- 7. Single cell measurements and analysis is needed to understand varying levels of heterogeneity 7 Single cell measurements reveal mutual exclusivity of subclonal mutations! cells Jean Fan | ISMB 2018 Scenario 1 (co-occurring) Scenario 2 (mutually exclusive) or
- 8. Single cell measurements and analysis is needed to understand varying levels of heterogeneity 8Jean Fan | ISMB 2018 Transcriptional Differences?
- 9. Jean Fan | ISMB 2018 9 Connecting Genetic and Transcriptional Heterogeneity at the Single-Cell Level
- 10. How to simultaneously assess transcriptional and genetic heterogeneity at the single cell level? 10Jean Fan | ISMB 2018 1. Technology Development Approach (Targeted RT-QPCR) 2. Computational Methods Approach (HoneyBADGER)
- 11. How to simultaneously assess transcriptional and genetic heterogeneity at the single cell level? 11Jean Fan | ISMB 2018 1. Technology Development Approach (Targeted RT-QPCR) 2. Computational Methods Approach (HoneyBADGER) bit.ly/GR_WangFan
- 12. Idea: If we can infer (some) genetic information from scRNA-seq, then we can directly connect genetic and transcriptional heterogeneity in single cells 12Jean Fan | ISMB 2018 scRNA-seq genetics
- 13. High mono-allelic detection and sparse coverage limits ability to confidently infer point mutations in scRNA-seq data Jean Fan | ISMB 2018 13
- 14. Joint analysis on multiple SNPs such as within CNVs enables genotypic classification of cells Jean Fan | ISMB 2018 14
- 15. HoneyBADGER identifies and assesses the status of CNVs in single cells using scRNA-seq data HMM-integrated Bayesian Approach to CNV Detection from single cell RNA-seq (HoneyBADGER) Jean Fan | ISMB 2018 15 Soo Lee Peter Park Woong-Yang ParkHae-Ock Lee jef.works/HoneyBADGER Just accepted! August 2018
- 16. HoneyBADGER identifies and assesses the status of CNVs in single cells using scRNA-seq data Jean Fan | ISMB 2018 16
- 17. 1-2) Single cells are clustered and pooled based on smoothed minor allele fractions Jean Fan | ISMB 2018 17
- 18. 3) HMM model identifies regions affected by CNVs and LOHs Jean Fan | ISMB 2018 18
- 19. 4-5) Bayesian model assesses posterior probability of each CNV in each cell Jean Fan | ISMB 2018 19
- 20. 6-7) Cells are split into branches and algorithm is recursively applied to each branch Jean Fan | ISMB 2018 20
- 21. HoneyBADGER separates tumor from normal cells, consistent with bulk tumor purity estimates Sample MM16 MM16-R Purity 81-95% 4.5-20% Ncells 23 21 Jean Fan | ISMB 2018 21
- 22. Jean Fan | ISMB 2018 22 HoneyBADGER separates tumor from normal cells, consistent with bulk tumor purity estimates
- 23. Jean Fan | ISMB 2018 23 HoneyBADGER separates tumor from normal cells, consistent with bulk tumor purity estimates
- 24. Confirm CNVs by WES and FISH Jean Fan | ISMB 2018 24
- 25. HoneyBADGER separates MM from normal cells based on identified CNVs using 3’ scRNA-seq data Jean Fan | ISMB 2018 25
- 26. HoneyBADGER separates MM from normal cells based on identified CNVs using 3’ scRNA-seq data Jean Fan | ISMB 2018 26
- 27. HoneyBADGER separates MM from normal cells based on identified CNVs using 3’ scRNA-seq data Jean Fan | ISMB 2018 27
- 28. HoneyBADGER separates MM from normal cells based on identified CNVs using 3’ scRNA-seq data Jean Fan | ISMB 2018 28
- 29. HoneyBADGER applied to progressive MM provides insights into subclonal expansion Jean Fan | ISMB 2018 29 MM34 (bone marrow) MM34A (ascites) Bulk WES
- 30. HoneyBADGER applied to progressive MM provides insights into subclonal expansion Jean Fan | ISMB 2018 30 Bulk WES scRNA-seqMM34 (bone marrow) MM34A (ascites)
- 31. HoneyBADGER identifies CNVs in MM34A Jean Fan | ISMB 2018 31
- 32. HoneyBADGER identifies CNVs in MM34A Jean Fan | ISMB 2018 32
- 33. HoneyBADGER identifies CNVs in MM34A Jean Fan | ISMB 2018 33
- 34. scRNA-seq data used by HoneyBADGER directly lends itself to transcriptomic comparison of subclones Jean Fan | ISMB 2018 34
- 35. scRNA-seq data used by HoneyBADGER directly lends itself to transcriptomic comparison of subclones GO:0007049 cell cycle Jean Fan | ISMB 2018 35
- 36. scRNA-seq data used by HoneyBADGER directly lends itself to transcriptomic comparison of subclones Jean Fan | ISMB 2018 36 GO:0002376 immune system processes
- 37. Alternatively characterize transcriptional heterogeneity and assess correspondence with genetic heterogeneity Jean Fan | ISMB 2018 37 bit.ly/NATURE_Fan
- 38. Alternatively characterize transcriptional heterogeneity and assess correspondence with genetic heterogeneity Jean Fan | ISMB 2018 38
- 39. Alternatively characterize transcriptional heterogeneity and assess correspondence with genetic heterogeneity Jean Fan | ISMB 2018 39 HoneyBADGER infers CNV from single-cell RNA-seq data to simultaneously assess gene expression and CNV status from same single cell
- 40. Jean Fan | ISMB 2018 40 Software and Tutorials Available at jef.works/HoneyBADGER
- 41. Summary ◦ Point mutation detection from scRNA-seq is difficult due to mono-allelic detection ◦ HoneyBADGER uses allele information across many SNPs to identify and infer the probability of CNVs and LOHs ◦ Inferring CNVs and LOHs from scRNA-seq allows direct comparison of genetic heterogeneity with transcriptional heterogeneity ◦ Integrative approaches at the single-cell level are needed to connect transcriptional and genetic heterogeneity in the context of a single patient / common microenvironment Jean Fan | ISMB 2018 41
- 42. Thanks and happy to take questions! Zhuang Lab Xiaowei Zhuang Jeff Moffitt Stephen Eichorn Guiping Wang Kharchenko Lab Peter Kharchenko Joseph Herman Nikolas Barkas Ruslan Soldatov Wu Lab Catherine Wu Lili Wang Ken Livak Shuqiang Li Jean Fan | ISMB 2018 42 Zhang Lab Kun Zhang Blue Lake Brandon Sos Song Chen Chun Lab Jerold Chun Gwen Kaeser Many others CZ Zhang Angela Brooks Alexander Misharin Rohan Verman Find me online! Web: http://JEF.works Github: JEFworks Twitter: @JEFworks jeanfan@fas.harvard.edu Park Lab Peter Park Soo Lee Semin Lee SGI Woong-yang Park Hae-Ock Lee Walsh Lab Chris Walsh Xiaochang Zhang Dulac Lab Catherine Dulac Eric Vaughn Mentors Collaborators Funding
Editor's Notes
- Working with Jan Burger and Catherine Wu, we looked at how genetic heterogeneity changed over time following treatment for 5 chronic lymphocytic leykemia patients with with bulk whole exome DNA sequencing Patient 1 was particularly remarkable We were able to collect samples prior to (Fludarabine, cyclophosphamide and rituximab)FCR chemotherapy Prior to the start of the targeted therapy ibrutinib After ibrutinic, At 1 2 year time point prior to relapse And during relapse
- Patient 1 was particularly remarkable because while we could see different mutations shift and change in their frequencies, by the last time point, we observed 4 different mutations iwthin PLCG2 Ibrutinib targets Bruton agammaglobulinemia tyrosine kinase (BTK) PLCG2 is the immediate downstream effector phospholipase
- From bulk DNA sequencing, we are not able to resolve whether mutations are co-occuring or mutually exclusive
- From targeted single cell mutation detection, we were able to confirm that these mutations affected different cells
- For that, we would need to connect aspects of heterogeneity
- In order to simultaneously assess
- In order to simultaneously assess
- The second way we attempted to address the integration of dna and rna information from the same single cell is through statistics Although we were not able to identify single nucleotide mutations from single cell RNA-seq with confidence we thought that joint consideration of many sites could achieve genotypic classification (again we see this pooling of information in order to improve signal and power) For example when we look at allelic patterns in neutral diploid regions, we can see a clear difference compared to deletion regions Here each column is a heterozygous SNP Each row is a cell Consider yellow as equal expression of both alleles Blue is expression of one allele And red is expression of the other So for a diploid region, even though an individual site may be mono-allelically detected, when we look across the neutral region, we are able to observe the other allele thus informing us that this cell indeed has both alleles
- So we sought to quantify this allelic imbalance to identify CNVs
- We developed a bayesian hierarchical model to assess the posterior probability a particular cell has a deletion given the observed heterozygous SNPs in a region of interest Of course, SNPs can be organized into genes, and since mono-allelic expression is a function at the gene level, the hierarchical model gave an intuitive way to organize our observations
- To identify these regions to test, we took a recursive hidden markov based approach To enhance signal, we first pool cells
- Use an HMM to identify regions to test
- Given these regions, we apply our full bayesian model
- In this manner we are able to sensitively identify alterations as small as 10mbs