Question: Can you give more information on the statistical test used for differential expression analysis in Cell Ranger?
Answer: As described in the "Differential Expression" section of the algorithms overview, Cell Ranger performs a variant of the negative binomial exact test based on the published sSeq method. When the counts become large, Cell Ranger switches to the fast asymptotic beta test used in edgeR.
The main difference between the sSeq approach and the other negative binomial variants described in the sSeq paper is in how the dispersion/variance is estimated. The supplementary pdf from the sSeq publication goes over the procedure for variance/dispersion estimation.
The tests generate p-values which are adjusted for multiple testing using the Benjamini-Hochberg procedure to control FDR.
The "Significant Genes" test in Loupe Cell Browser replicates the "Differential Expression" analysis in Cell Ranger.
To explore the code for differential expression analysis on your own, please see our Github repository here. Disclaimer: Source code for 10x pipelines is available at https://github.com/10XGenomics for interested customers. However, 10X Genomics is not staffed to answer questions about it or provide support for modifications.