Topic
TPM vs counts for differential expression
Can I run DESeq2 on TPM or FPKM values?
Count-based tests model the sampling variance of integer counts. TPM and FPKM have already divided that information away, so feeding them into DESeq2 or edgeR produces wrong dispersion estimates and wrong p-values. Use raw counts, or tximport with length offsets when you started from salmon or kallisto.
- Why You Must Not Use TPM for Differential Expression in Bulk RNA-seq
TPM already threw away the count magnitude DESeq2 and edgeR need to estimate dispersion, and rounding it to dodge the integer check does not bring that information back.