LN's $t$-test

Differential expression testing for single-cell transcriptomics with reduced false discoveries.

Installation

pip install ln-ttest

Usage

from lntest import rank_genes_groups_ln

rank_genes_groups_ln(adata, groupby="leiden", layer="norm_counts") # normalized, not log-transformed
adata.uns["rank_genes_groups"]["logfoldchanges"]
adata.uns["rank_genes_groups"]["lfc_se"]

To select each cluster's markers, we recommend taking the genes LN's $t$-test calls significant and up in the cluster, ranked by the log ratio of their detection rates over its standard error:

rank_genes_groups_ln(adata, groupby="leiden", layer="norm_counts", corr_method="bonferroni", rankby="markers")
adata.uns["rank_genes_groups"]["names"]["3"][:20]  # cluster 3's top 20 markers

Example

The notebook PBMC3k showcases LN's $t$-test finding far fewer DEGs than scanpy's $t$-test on the 3k PBMCS data set from 10x Genomics.