where fi,j is the fraction of non-zero expression values in the cluster and fi,j¯ is the fraction of non-zero expression values for cells not in the cluster. Similarly, μi,j is the mean expression in the cluster and μi,j¯ is the mean expression for cells not in the cluster. Small constants ε1=0.1 and ε2=0.01 are added to prevent the enrichment score from going to infinity as the mean or non-zero fractions go to zero. Enrichment scores are available as matrix layer enrichment in the aggregated Loom files (named “…agg.loom”). We also computed an enrichment q value by shuffling the expression matrix, available as layer enrichment_q. To find genes enriched at a 10% false discovery rate, for example, simply select genes with q scores below 0.1.
