Sequencing Coverage Calculator

Calculate sequencing coverage, expected reads, and genome assembly quality estimates.

Genome or target size (Mb) (Mb)

Human: 3000Mb, E.coli: 4.6Mb

Total reads (reads)

Read length (bp) (bp)

Insert size for paired-end (bp) (bp)

Please check your inputs and try again.

Coverage Is an Average, Not a Guarantee

Coverage is total sequenced bases divided by genome size. That average conceals enormous variation, because reads land according to Poisson statistics rather than evenly.

Coverage = (reads × read length) / genome size

At 10× average coverage, roughly 0.005% of positions receive no reads at all by chance alone. Across a 3 Gb genome that is around 136,000 uncovered bases. This is the Lander–Waterman result, and it is why 30× is standard for whole genome sequencing rather than the 10× that naive arithmetic might suggest.

Application	Typical depth	Why
Whole genome, germline	30×	Confident diploid genotype calls
Whole exome	100×	Uneven capture efficiency
Somatic tumour	100–500×	Low-frequency subclonal variants
RNA-seq differential expression	20–30M reads	Depth measured in reads, not fold
Bacterial assembly	50–100×	Cheap; improves contiguity

Why Heterozygotes Need More Depth

Calling a homozygous variant needs only a few reads agreeing. A heterozygote requires enough reads to see both alleles — and allelic sampling is binomial, so at low depth one allele can be missed entirely.

At 10× coverage there is roughly a 0.2% chance of drawing all ten reads from one allele of a true heterozygote, producing a false homozygous call. At 30× that probability becomes negligible, which is the real justification for the standard.

Uneven Coverage in Practice

Cause	Effect
GC-rich or GC-poor regions	Reduced amplification and capture
Repetitive sequence	Reads map ambiguously or not at all
Exome capture variability	Bait efficiency differs several-fold between targets
PCR duplicates	Inflate apparent depth without adding information

Duplicates matter particularly: they are copies of the same original fragment, so they raise the coverage number without contributing independent evidence. Deduplication before variant calling is standard, and effective coverage after it is often noticeably lower than the raw figure.

Worked Examples

Example 1: Human WGS: 3000Mb, 600M reads×150bp

Coverage=6e8×150/3e9

Result: Coverage=30× — standard WGS depth

30× gives >99.9% genome coverage

Example 2: E.coli assembly: 4.6Mb, 1M reads×150bp

Coverage=1e6×150/4.6e6

Result: Coverage=32.6× — excellent bacterial WGS

Enough for de novo assembly with long reads

Example 3: Why 30× not 10×

Heterozygous site at 10× depth

Result: About 0.2% chance of a false homozygous call

All ten reads can come from one allele by chance. At 30× the binomial probability becomes negligible, which is the basis of the standard.

Example 4: Somatic variant detection

Subclonal variant at 5% allele frequency

Result: Needs 100–500×

At 30× a 5% variant is expected in only 1–2 reads, indistinguishable from sequencing error. Deep coverage is essential for tumour work.

Common Mistakes

⚠️

Treating coverage as uniform

Reads land by Poisson statistics. At 10× average, tens of thousands of positions in a human genome receive no reads at all.

⚠️

Using 10× for germline variant calling

Heterozygotes can be missed when allelic sampling happens to favour one allele. Thirty-fold is standard for this reason.

⚠️

Counting PCR duplicates as coverage

Duplicates are copies of the same fragment and add no independent evidence. Effective coverage after deduplication is often substantially lower.

⚠️

Applying fold-coverage thinking to RNA-seq

Transcript abundance varies over orders of magnitude, so RNA-seq depth is specified in reads per sample rather than fold coverage.

Frequently Asked Questions

Lander-Waterman model?

C=NL/G where C=coverage, N=reads, L=read length, G=genome size. Poisson model: P(0 reads) = e^-C. At 30×: 99.997% bases covered. Gaps occur at repetitive regions regardless of depth.

Read depth for different applications?

WGS germline: 30×. Cancer somatic (tumor purity): 60-100×. cfDNA liquid biopsy: >1000×. RNA-seq: 20-50M reads. ChIP-seq: 20-50M reads. ATAC-seq: 50-100M reads. Single-cell: 10,000-50,000 reads/cell.

Why is 30× standard for human genomes?

Because heterozygous sites need enough reads to observe both alleles. At 10× there is a meaningful chance of sampling only one and making a false homozygous call.

What is the Lander–Waterman model?

It describes read distribution as a Poisson process, predicting what fraction of the genome receives no coverage at a given average depth.

Why do tumour samples need deeper coverage?

Somatic variants may be present in a small subclone. Detecting a 5% allele fraction reliably requires 100–500× depth.

Do PCR duplicates count toward coverage?

No. They are copies of the same original fragment and provide no independent evidence, so they should be removed before variant calling.