1 Differential Gene Expression Test

We can compare the gene expression profiles of cells under each KO (gRNA) condition with those of cells without this gRNA using either Wilcoxon rank sum test or Welch’s t-test.

We have 2 variations on the gene expression data we use: (1) log-CPM expression; (2) corrected expression: log-CPM data corrected for batch effect and other confounding factors and then scaled. The 2nd kind is the expression data we passed to GSFA.

2 LFSR Estimation from GSFA (NTC condition regressed out, singletons)

In terms of our GSFA method, we can also determine if a KO condition has an significant effect on gene expression in the following way:

For a given GSFA inference result, we estimated the local false sign rate (lfsr) of a certain knock-out effect on each gene based on the posteriors of \(\beta\) and \(W\).

For gene \(j\) and KO condition \(m\), \[lfsr_{mj} = \text{min} \Big\{\text{Pr}(\sum_{k=1}^K \beta_{mk}W_{jk} \geq 0 \text{ | Data}), \text{Pr}(\sum_{k=1}^K \beta_{mk}W_{jk} \leq 0 \text{ | Data}) \Big\}\] Under each KO condition, the number of genes that pass a certain LFSR cutoff would be compared to that from a the DGE tests mentioned above.