Why average coverage doesn’t tell the whole story
A “30× genome” sounds fully covered. But an average is just an average — and because reads land at random, even good average depth leaves some bases with no coverage at all.
Average depth is only a mean
Coverage depth is the average number of reads spanning each base — total sequenced bases divided by genome size. (The coverage explainer covers how to plan that number.) The catch is that it is a single summary of a whole distribution: some positions get more reads than the average, and some get fewer.
Randomness alone creates gaps
Reads land at effectively random positions across the genome. With randomness, some regions simply get unlucky and receive few or zero reads, even when the average is healthy. Coverage is a distribution, not a guarantee — so “30× on average” does not mean “every base has about 30 reads.”
The relationship is exponential, not linear
The Lander–Waterman model captures this: at an average depth of C, the chance that any given base is missed entirely is e−C. Because it is exponential, raising depth shrinks the gap fraction dramatically rather than proportionally. The expected fraction of genome left uncovered works out to:
5× → ≈ 0.67% uncovered
15× → ≈ 0.00003% uncovered
30× → ≈ 10⁻¹¹% uncovered (effectively none)
Notice the jump from 5× to 15× is not three times better — it is roughly twenty thousand times better. Even going from 5× to 10× takes the uncovered fraction from about 0.67% down to about 0.0045%, a 150-fold drop, not a halving.
Why this drives depth targets
This is why projects that need near-complete coverage — clinical whole-genome sequencing, for instance — target higher average depth than the bare “reads per base” arithmetic might suggest. The extra depth is not about the average; it is about pushing the unlucky tail of the distribution down so few bases are left uncovered. (And real data is gappier still, because coverage is not perfectly random — GC bias, repeats and mappability make some regions systematically harder.)