Journal Club
Comparing methods for differential expression
viktor.petukhov@pm.me
Viktor Petukhov,
Khodosevich group
Confronting false discoveries in single-cell differential expression
bioRxiv 2021.03.12.435024;
Jordan W. Squair, Matthieu Gautier, Claudia Kathe, Mark A. Anderson, Nicholas D. James, Thomas H. Hutson, Rémi Hudelle, Taha Qaiser, Kaya J. E. Matson, Quentin Barraud, Ariel J. Levine, Gioele La Manno, Michael A. Skinnider, Grégoire Courtine
"Ground-truth"
18 datasets
- matched bulk and scRNA-seq
- performed on the same population of purified cells
- exposed to the same perturbations
- sequenced in the same laboratories
Evaluated methods
A curated database reveals trends in single cell transcriptomics.
Valentine Svensson, Eduardo da Veiga Beltrame, & Lior Pachter
Evaluated methods
- Only DE across conditions, no DE between types
- 8 single-cell methods
- 6 pseudo-bulk methods
Recovering bulk results
DE
GO
cells stimulated with poly-I:C
What's wrong with single-cell methods?
Bias towards highly-expressed genes
Spike-in experiments
Correlation between expression and -log10(p-value)
False-positives
What causes this bias?
Shuffled aggregation
No aggregation
What causes this bias?
Shuffled aggregation
No aggregation
What causes this bias?
False-positives on simulated data
False-positives on simulated data
Human vs mouse data
∆(variance) = variance in biological replicates minus variance in pseudo-replicates
Conclusions
- Single-cell-based methods fail to capture inter-individual variability and should not be used
- edge-R is the most stable
- It's important to control for covariates in biological experiments
A practical solution to pseudoreplication bias in single-cell studies.
Nat Commun 12, 738 (2021).
Zimmerman, K.D., Espeland, M.A. & Langefeld, C.D.
Simulating false-positive rates
Simulating sensitivity
Conclusions
Single-cell-based methods fail even on a theoretical level