Additional reading

/5 - Olga's paper rating on a scale of 1 (worst) to 5 (best)

4/5 Brennecke et al Nature Methods (2013)
- Frequentist based approach of assigning an appropriate gene expression threshold where you can trust the differences
4/5 Kharchenko et al Nature Methods (2014)
- Bayesian based approach of finding genes that are differentially expressed based on how well that gene was captured across all cells (the prior)

5/5 COMBAT
- Adjust for batches in the data
- This has >1000 citations!!
5/5 Hicks et al
- Exposé meta-analysis of almost all published single-cell RNA seq papers so far and shows how many of them have confounded their biological signal with technical batches
4/5 SVA
- Can specify that you want to correct for something (like RIN) but don't correct for what you're interested in. But... often in single cell data you're trying to find new populations so you don't know a prior what you want to not be corrected for
4/5 RUVseq
- With the "RUVg" version can specify a set of control genes that you know aren't supposed to change between groups (maybe from a bulk experiment) but they say in their manual not to use the normalized counts for differential expression, only for exploration, because you may have corrected for something you actually DID want but didn't know
4/5 scLVM
- This method claims to account for differences in cell cycle stage and help to put all cells onto the same scale, so you can then do pseudotime ordering and clustering and all that jazz.