Towards progress in batch effects and biases in single-cell RNA-Seq data

Presented at the 2016 Single Cell Genomics Conference at the Wellcome Genome Campus in Hinxton, UK

Transcript

1. Towards progress in
   batch effects and biases in
   single-cell RNA-Seq data
   Stephanie Hicks
   Dana-Farber Cancer Institute / Harvard SPH
   @stephaniehicks
   Single Cell Genomics 2016
   Wellcome Genome Campus
   

   View Slide

2. Bimodal distribu=on of expression
   Number of Cells
   Expression level of Gene X
   μ
   1
   μ
   2
   True Expression
   

   View Slide

3. Bimodal distribu=on of expression
   Expression level of Gene X
   μ
   1
   μ
   2
   True Expression
   Observed Expression
   Number of Cells
   

   View Slide

4. Patel et al. (2014) Science
   Cells cluster by tumor
   

   View Slide

5. Verhaak et al. (2010). Cancer Cell
   

   View Slide

6. Leek et al. (2010) Nat Reviews Genetics
   Batch effects in genomics data
   

   View Slide

7. Processing*Batch*
   Biological*Group*
   Completely*confounded*study*design*
   Balanced*study*design*
   Rep:*1* Rep:*2* Rep:*2* Rep:*2*
   Rep:*1* Rep:*1*
   Group:*1*
   Group:*2*
   Group:*3*
   Batch:*3* Batch:*1*
   Batch:*2*
   Batch:*3*
   Observed*Differences*
   We*cannot*determine*if*
   variaCon*is*driven*by*
   biology*or*batch*effects*
   ●
   ●
   Batch
   Batch 1
   Batch 2
   Batch 3
   ●
   ●
   Batch
   Batch 1
   Batch 2
   Batch 3
   ●
   ●
   ●
   ●
   ●
   ●
   Batch
   Batch 1
   Batch 2
   Batch 3
   Batch:*2* Batch:*1*
   Group:*3* Group:*1*
   Group:*2*
   The$Problem$of$Confounding$Biological$Varia6on$and$Batch$Effects$
   ProporCon*of*detected*genes*
   ProporCon*of*detected*genes*
   Principal*Component*2*
   Principal*Component*2*
   Principal*Component*1*
   Principal*Component*1*
   Principal*Component*2*
   Principal*Component*1*
   ProporCon*of*detected*genes*
   Group*1*
   Group*1* Group*2* Group*3*
   Group*2* Group*3*
   Group*1* Group*2* Group*3*
   Good*
   Bad*
   When cells from one biological group or
   condition are cultured, captured and
   sequenced separate from cells in a
   second condition
   Confounded study design
   Batch effects in
   single-cell RNA-Seq data
   

   View Slide

8. Processing*Batch*
   Biological*Group*
   Completely*confounded*study*design*
   Balanced*study*design*
   Rep:*1* Rep:*2* Rep:*2* Rep:*2*
   Rep:*1* Rep:*1*
   Group:*1*
   Group:*2*
   Group:*3*
   Batch:*3* Batch:*1*
   Batch:*2*
   Batch:*3*
   Observed*Differences*
   We*cannot*determine*if*
   variaCon*is*driven*by*
   biology*or*batch*effects*
   ●
   ●
   Batch
   Batch 1
   Batch 2
   Batch 3
   ●
   ●
   Batch
   Batch 1
   Batch 2
   Batch 3
   ●
   ●
   ●
   ●
   ●
   ●
   Batch
   Batch 1
   Batch 2
   Batch 3
   Batch:*2* Batch:*1*
   Group:*3* Group:*1*
   Group:*2*
   The$Problem$of$Confounding$Biological$Varia6on$and$Batch$Effects$
   ProporCon*of*detected*genes*
   ProporCon*of*detected*genes*
   Principal*Component*2*
   Principal*Component*2*
   Principal*Component*1*
   Principal*Component*1*
   Principal*Component*2*
   Principal*Component*1*
   ProporCon*of*detected*genes*
   Group*1*
   Group*1* Group*2* Group*3*
   Group*2* Group*3*
   Group*1* Group*2* Group*3*
   Good*
   Bad*
   When cells from one biological group or
   condition are cultured, captured and
   sequenced separate from cells in a
   second condition
   More balanced study design:
   Cells from different biological group
   processed in same batch
   Batch effects in
   single-cell RNA-Seq data
   

   View Slide

9. Processing*Batch*
   Biological*Group*
   Completely*confounded*study*design*
   Balanced*study*design*
   Rep:*1* Rep:*2* Rep:*2* Rep:*2*
   Rep:*1* Rep:*1*
   Group:*1*
   Group:*2*
   Group:*3*
   Batch:*3* Batch:*1*
   Batch:*2*
   Batch:*3*
   Observed*Differences*
   We*cannot*determine*if*
   variaCon*is*driven*by*
   biology*or*batch*effects*
   ●
   ●
   Batch
   Batch 1
   Batch 2
   Batch 3
   ●
   ●
   Batch
   Batch 1
   Batch 2
   Batch 3
   ●
   ●
   ●
   ●
   ●
   ●
   Batch
   Batch 1
   Batch 2
   Batch 3
   Batch:*2* Batch:*1*
   Group:*3* Group:*1*
   Group:*2*
   The$Problem$of$Confounding$Biological$Varia6on$and$Batch$Effects$
   ProporCon*of*detected*genes*
   ProporCon*of*detected*genes*
   Principal*Component*2*
   Principal*Component*2*
   Principal*Component*1*
   Principal*Component*1*
   Principal*Component*2*
   Principal*Component*1*
   ProporCon*of*detected*genes*
   Group*1*
   Group*1* Group*2* Group*3*
   Group*2* Group*3*
   Group*1* Group*2* Group*3*
   Good*
   Bad*
   When cells from one biological group or
   condition are cultured, captured and
   sequenced separate from cells in a
   second condition
   More balanced study design:
   Cells from different biological group
   processed in same batch
   Batch effects in
   single-cell RNA-Seq data
   Proposed by
   Tung et al.
   (2016) bioRxiv
   

   View Slide

10. Processing*Batch*
    Biological*Group*
    Completely*confounded*study*design*
    Balanced*study*design*
    Rep:*1* Rep:*2* Rep:*2* Rep:*2*
    Rep:*1* Rep:*1*
    Group:*1*
    Group:*2*
    Group:*3*
    Batch:*3* Batch:*1*
    Batch:*2*
    Batch:*3*
    Observed*Differences*
    We*cannot*determine*if*
    variaCon*is*driven*by*
    biology*or*batch*effects*
    ●
    ●
    Batch
    Batch 1
    Batch 2
    Batch 3
    ●
    ●
    Batch
    Batch 1
    Batch 2
    Batch 3
    ●
    ●
    ●
    ●
    ●
    ●
    Batch
    Batch 1
    Batch 2
    Batch 3
    Batch:*2* Batch:*1*
    Group:*3* Group:*1*
    Group:*2*
    The$Problem$of$Confounding$Biological$Varia6on$and$Batch$Effects$
    ProporCon*of*detected*genes*
    ProporCon*of*detected*genes*
    Principal*Component*2*
    Principal*Component*2*
    Principal*Component*1*
    Principal*Component*1*
    Principal*Component*2*
    Principal*Component*1*
    ProporCon*of*detected*genes*
    Group*1*
    Group*1* Group*2* Group*3*
    Group*2* Group*3*
    Group*1* Group*2* Group*3*
    Good*
    Bad*
    When cells from one biological group or
    condition are cultured, captured and
    sequenced separate from cells in a
    second condition
    More balanced study design:
    Cells from different biological group
    processed in same batch
    Batch effects in
    single-cell RNA-Seq data
    Implemented by Tung et al. (2016) bioRxiv
    

    View Slide

11. Use FASTQ header as a surrogate for batch
    http://support.illumina.com/help/SequencingAnalysisWorkflow/Content/Vault/Informatics/Sequencing_Analysis/CASAVA/
    swSEQ_mCA_FASTQFiles.htm
    

    View Slide

12. Processing*Batch*
    Biological*Group*
    Completely*confounded*study*design*
    Balanced*study*design*
    Rep:*1* Rep:*2* Rep:*2* Rep:*2*
    Rep:*1* Rep:*1*
    Group:*1*
    Group:*2*
    Group:*3*
    Batch:*3* Batch:*1*
    Batch:*2*
    Batch:*3*
    Observed*Differences*
    We*cannot*determine*if*
    variaCon*is*driven*by*
    biology*or*batch*effects*
    ●
    ●
    Batch
    Batch 1
    Batch 2
    Batch 3
    ●
    ●
    Batch
    Batch 1
    Batch 2
    Batch 3
    ●
    ●
    ●
    ●
    ●
    ●
    Batch
    Batch 1
    Batch 2
    Batch 3
    Batch:*2* Batch:*1*
    Group:*3* Group:*1*
    Group:*2*
    The$Problem$of$Confounding$Biological$Varia6on$and$Batch$Effects$
    ProporCon*of*detected*genes*
    ProporCon*of*detected*genes*
    Principal*Component*2*
    Principal*Component*2*
    Principal*Component*1*
    Principal*Component*1*
    Principal*Component*2*
    Principal*Component*1*
    ProporCon*of*detected*genes*
    Group*1*
    Group*1* Group*2* Group*3*
    Group*2* Group*3*
    Group*1* Group*2* Group*3*
    Good*
    Bad*
    Hicks et al. (2015) bioRxiv
    

    View Slide

13. Cells cluster by tumor
    Hicks et al. (2015) bioRxiv
    

    View Slide

14. Cells cluster by batch
    (tumors are confounded with batch)
    Hicks et al. (2015) bioRxiv
    

    View Slide

15. Hicks et al. (2015) bioRxiv
    

    View Slide

16. Different batches have different detec=on rates
    Hicks et al. (2015) bioRxiv
    

    View Slide

17. Deng et al. (2014) Science
    Treutlein et al. (2014) Nature
    Trapnell et al. (2014) Nat Biotech
    Patel et al. (2014) Science
    

    View Slide

18. Batch and outcomes of interest are confounded in
    published scRNA-Seq experiments
    

    View Slide

19. Differences in the detec=on rates
    between batches of cells
    Study design #1:
    

    View Slide

20. Study design #2:
    Differences in the detec=on rates
    between batches of cells
    

    View Slide

21. Two studies had completely confounded study designs
    

    View Slide

22. Correla=on between PC1 and
    propor=on of detected genes
    Propor=on of detected genes
    Shalek et al. (2014). Nature
    Finak et al. (2015). Genome Biology
    

    View Slide

23. View Slide

24. View Slide

25. View Slide

26. View Slide

27. View Slide

28. View Slide

29. View Slide

30. View Slide

31. View Slide

32. View Slide

33. View Slide

34. View Slide