Planning RNA-seq read depth (and why it isn’t RPKM)

RNA-seq depth is measured in reads per sample, not genome coverage, because transcript abundance spans orders of magnitude. A simple estimate gets you a planning figure; the skewed expression distribution is what makes it only an estimate.

Reads per sample, not ×coverage

You do not sequence a transcriptome to a uniform depth the way you do a genome — a handful of highly expressed genes soak up a large share of reads while most genes get few. So RNA-seq is planned as a target number of reads per sample. A back-of-envelope figure:

reads ≈ (number of features × target reads per feature) ÷ usable fraction

For ~20,000 human genes at an average of 100 reads each, with 80% of reads usable after rRNA, multi-mapping and QC losses: (20,000 × 100) ÷ 0.8 =2.5 M reads as a floor. In practice many labs target 20–30 M reads per sample for differential expression, i.e. a much higher average per gene, precisely because of the skew.

Why the average lies

That formula assumes every gene gets the average. Real expression is long-tailed: low-expression genes may get a handful of reads even when the mean is comfortable, and those are exactly the genes where you lose power to detect change. The estimate is an order-of-magnitude planning number, not a statistical power calculation — for that you need a tool that models dispersion and the expression distribution.

Depth is not RPKM

Sequencing depth (reads per sample) is what you order; RPKM/FPKM/TPM arenormalised expression units you compute afterwards to compare genes within and across samples. To calculate RPKM for a gene:RPKM = (reads mapped to the gene × 10⁹) ÷ (total mapped reads × gene length in bp)— it divides a gene’s read count by its length and the library size, so it is a per-gene measure derived from the data, not a knob you set before the run. Plan in reads; report in TPM.