How much depth do you need to call a variant?

“More depth is better” is not a plan. How much you actually need depends on how rare the variant is — its allele frequency — far more than on any blanket coverage number.

Allele frequency sets the difficulty

The variant allele frequency (VAF) is the fraction of the molecules at a position that carry the variant. A germline heterozygous variant sits at about 50% VAF — half the reads should carry it — so it is easy to see. A subclonal somatic mutation, or a tumour fragment in circulating tumour DNA (ctDNA), might be present at 1% VAF or lower. The rarer it is, the more reads you must collect before you expect to catch even a few copies of it.

Why it’s a sampling problem

Each read at a position is like one random draw from a pool that is mostly reference and occasionally variant. Calling a variant means catching enough of those rare draws — say at least three reads supporting it — to be confident it is real and not a sequencing error. At 50% VAF almost every handful of reads contains some; at 1% VAF you need roughly a hundred reads just to expect a single one, and several times that to be confident of catching three. This is the binomial sampling intuition; the calculator turns it into an exact probability so you don’t have to work the formula by hand.

A concrete contrast

At 30× depth, a 50% VAF germline variant gives about 15 supporting reads on average — a confident call. The same 30× on a 1% VAF somatic variant gives about 0.3 supporting reads on average, i.e. usually none: you would miss it. To have a good chance of seeing ≥3 reads of a 1% variant you typically need hundreds to over a thousand× — which is why low-VAF work looks so different from germline sequencing.

Typical depth targets in practice

Treat these as typical ranges practitioners use, not hard rules — the right number follows from your target VAF and how many supporting reads your caller needs.