Clusters and/or cell types have been identified, we now want to compare sample groups:

• Differential expression - Differences in expression between sample group within a biological state.

• Differential abundance - Differences in cell numbers between sample groups within a biological state.

Replicates are samples not cells:

• Single cells within a sample are not independent of each other
• Using cells as replicates amounts to studying variation inside an individual
• May violate assumptions of independence and lead to inflated false positives
• We want to study variation across a population

Are the genes up or down regulated between treated vs control or wildtype vs mutant or healthy vs diseased etc. ?

• Create pseudo-bulk samples by summing raw counts across cells for each sample

Once the pseudo-bulk matrix is generated, we can use any bulk RNA-seq DE method to perform the analysis, such as DESeq2, edgeR, limma-voom etc.

We should remove pseudosamples with created from very few cells eg. < 20 cells

We should remove genes that are lowly expressed * reduces computational work, * improves the accuracy of mean-variance trend modelling * decreases the severity of the multiple testing correction * filter: log-CPM threshold in a minimum number of samples, smallest sample group

Seurat function FindMarkers() can also be used for differential expression.

The default Wilcoxon Rank Sum test was used to find cluster markers but the function is also a wrapper for several other methods, including MAST, DESeq2 and limma.

We can choose the method with the test.use parameter.

We will use DESeq2

DESeq2 will ‘normalise’ our pseudocount data to account for composition biases and differences in sequencing depth between samples.

Main Steps:

1. Estimate size factors to account for differences in sequencing depth between samples
2. Estimate dispersion parameters for each gene to model the variance of counts across samples
3. Fit a negative binomial generalized linear model (NB GLM) for each gene to
   • model the relationship between gene expression and experimental conditions
   • Use the Wald test to determine if our fold change is significant
   • run a B-H Multiple testing correction to control the false discovery rate (FDR)