Misuse of RPKM or TPM normalization when comparing across samples and sequencing protocols

Abstract

In recent years, RNA-sequencing (RNA-seq) has emerged as a powerful technology for transcriptome profiling. For a given gene, the number of mapped reads is not only dependent on its expression level and gene length, but also the sequencing depth. To normalize these dependencies, RPKM (reads per kilobase of transcript per million reads mapped) and TPM (transcripts per million) are used to measure gene or transcript expression levels. A common misconception is that RPKM and TPM values are already normalized, and thus should be comparable across samples or RNA-seq projects. However, RPKM and TPM represent the relative abundance of a transcript among a population of sequenced transcripts, and therefore depend on the composition of the RNA population in a sample. Quite often, it is reasonable to assume that total RNA concentration and distributions are very close across compared samples. Nevertheless, the sequenced RNA repertoires may differ significantly under different experimental conditions and/or across sequencing protocols; thus, the proportion of gene expression is not directly comparable in such cases. In this review, we illustrate typical scenarios in which RPKM and TPM are misused, unintentionally, and hope to raise scientists’ awareness of this issue when comparing them across samples or different sequencing protocols.