DNA Sequencing Coverage Calculator

Introduction: What Sequencing Depth and Coverage Breadth Actually Measure

Every sequencing plan starts with the same question: how much data is enough? "Coverage" answers it in two different ways, and confusing the two is the most common planning error in a core facility. Depth of coverage is an average: it is the number of times a typical base in your reference is read, written as 30×, 100× or 500×. Breadth of coverage is a proportion: it is the share of the reference that is seen at least once, or at least some threshold number of times. A run can have a comfortable average depth and still leave thousands of bases untouched, and a run with excellent breadth at 1× can be far too shallow to call a heterozygous variant with confidence.

This DNA sequencing coverage calculator computes both from the numbers you already have on a quote or a run-planning sheet: genome or capture-target size, read length, the number of reads or read pairs the flow cell will produce, whether the run is paired-end, how many reads survive trimming, quality filtering and mapping, and what fraction of the library is duplicated. It then reports mean depth, expected breadth, the fraction of bases at or above your target depth, the number of reads you would need to hit that target, and the cost of the output at your quoted price per gigabase. The depth relationship and the breadth model both come from the classic Lander and Waterman treatment of random shotgun sequencing, which is still the reference framework used in modern coverage guidance.

How to Use This DNA Sequencing Coverage Calculator

1. Enter the reference size and pick the matching unit. Use Gbp for whole mammalian genomes, Mbp for microbial genomes and exome capture space, Kbp for plasmids and amplicons. For a capture panel or an exome, enter the size of the target space, not the whole genome, or the depth will be badly understated.
2. Enter the read length in bases for a single read: 150 for a standard Illumina 2 × 150 run, 300 for a MiSeq 2 × 300 run, or the mean read length for a long-read platform.
3. Enter the number of reads or read pairs and choose whether that figure is a raw count, millions or billions. Instrument specifications are usually quoted in millions of reads or millions of clusters, so the unit selector prevents the most damaging unit mistake on this page.
4. Tick paired-end only if the number you entered counts fragments (read pairs). Ticking it multiplies read length by two, because both ends of the fragment are sequenced. If your number already counts individual reads, leave it unticked.
5. Set usable reads to the percentage that survives adapter trimming, quality filtering, mapping and (for capture work) on-target selection. Set the duplicate rate to the fraction that duplicate-marking will discard. The two are applied multiplicatively, so 92 percent usable with 8 percent duplicates leaves 84.6 percent of the raw bases.
6. Set the target depth you need per base. The calculator reports the expected share of bases at or above that depth and the read count required to reach it as a mean.
7. Optionally enter a cost per gigabase of raw output to price the run, then press Calculate Coverage. Re-run with a second scenario to see how depth, breadth and cost trade off before you commit samples.

The Lander-Waterman Coverage Formula, Term by Term

Shotgun sequencing scatters reads across the reference. If read start positions are approximately uniform and independent, the expected number of times a given base is read is the total number of sequenced bases divided by the size of the reference. That is the Lander-Waterman depth relationship:

Formula: C = (N × L) / G

Here is mean depth of coverage, is the number of usable reads, is read length in bases and is the reference size in bases. Rearranged, the total amount of sequence you must buy for a target depth is simply

Formula: B = G × C

so a 3.1 Gbp human genome at 30× needs 93 gigabases of usable aligned sequence, and rather more raw output. The calculator applies losses before the division. If is the usable-read percentage and is the duplicate percentage, the usable fraction is

Formula: f = q / 100 × (1 − d / 100)

and mean depth becomes , where is 2 for paired-end fragments and 1 otherwise. Inverting that expression gives the read count you need for a target depth :

Formula: N_req = (k × G) / (L × r × f)

Breadth of Coverage and the Poisson Gap Model

Lander and Waterman also give the piece most calculators leave out. If read starts are random and independent, the number of reads covering any particular base follows a Poisson distribution with mean . The probability that a base is missed entirely is therefore

Formula: P (0) = e^−C

and the expected breadth of coverage at one read or deeper is one minus that quantity:

Formula: breadth = 1 − e^−C

This single expression explains a great deal of practical sequencing behaviour. At 1× mean depth you expect to miss 36.8 percent of the reference. At 3× you still miss about 5 percent. At 5× the expected gap fraction is 0.67 percent, and only around 10× does the random-gap term become negligible next to real biological and technical bias. That is why "1× coverage" never means "the whole genome once".

The same Poisson model answers the question that actually matters for variant calling, namely the share of bases that reach a usable threshold. The calculator evaluates the upper tail

Formula: P (X ≥ k) = 1 − ∑ i = 0 k − 1 (e^−C C^i) / (i !)

using a normal approximation with a continuity correction when the depth is large enough that the direct sum would be slow or numerically awkward. The result is the headline number in most sequencing contracts: "at least 95 percent of the target at 20× or better". Notice that a mean of exactly 30× does not put 95 percent of bases at 30× or better; it puts roughly 71 percent there. Planning to a percentile always costs more than planning to a mean.

Usable Reads, Duplicates and the Gap Between Raw and Effective Output

The number printed on an instrument specification sheet is raw output. What reaches a variant caller is smaller, often by a fifth or more. Adapter read-through and quality trimming shorten reads. Reads that fail to map, map ambiguously or map outside a capture target are removed. PCR and optical duplicates are copies of a molecule that has already been observed, so they inflate the base count without adding independent evidence and are discarded by duplicate marking. Low-complexity libraries, over-amplified PCR products and over-clustered flow cells all push the duplicate rate up.

Because these losses stack, the calculator applies them multiplicatively rather than adding them, and reports both the raw output and the usable output so you can see the gap. If your quoted price is per gigabase of raw output, the effective price per usable gigabase is the number that belongs in the budget:

Formula: c_eff = c_raw / f

Worked Example: A 30× Human Whole-Genome Run

Take a human whole-genome resequencing run. Reference size 3.1 Gbp, read length 150 bases, 400 million read pairs, paired-end, 92 percent usable reads after trimming and mapping, an 8 percent duplicate rate, and a 30× target depth at a quoted price of six US dollars per raw gigabase. Entering exactly those values gives:

• Usable fraction: 0.92 × 0.92 = 0.8464.
• Raw output: 150 × 2 × 400,000,000 = 120.00 Gb.
• Usable output: 120.00 Gb × 0.8464 = 101.57 Gb.
• Mean depth: 101,568,000,000 ÷ 3,100,000,000 = 32.8×.
• Breadth at 1× or deeper: 1 − e^−32.76 = 100.0000 percent; fewer than one base is expected to be missed by chance alone.
• Bases at 30× or deeper: 70.90 percent — comfortably above the mean target, but well short of a 95 percent contractual threshold.
• Read pairs required for a 30× mean: 366,257,089.
• Cost: 120 Gb × $6 = $720.00, which is $7.09 per usable gigabase.

The instructive part is the last two bullets. The run clears its mean-depth target with room to spare, yet only about seven bases in ten reach 30×. If the study protocol demands 95 percent of bases at 30×, the Poisson tail says the mean must rise to roughly 41×, which is about a quarter more sequencing than the naive calculation suggests. That difference is exactly the sort of thing a coverage calculator should surface before the order is placed rather than after the BAM files are delivered.

Coverage Planning Table for Common Sequencing Designs

The table below is generated from the same formulas the calculator uses, so every row can be reproduced by typing the inputs into the form above. It is a quick sanity check on whether a proposed design is in the right order of magnitude.

Choosing a Target Depth for Your Application

Depth targets are set by the smallest signal you must detect, not by the size of the genome. Human germline variant calling from short reads is conventionally planned at 30× mean depth, because that is roughly where heterozygous single-nucleotide variant sensitivity plateaus. Somatic work in tumour samples routinely runs from 100× to several hundred × because a subclonal allele present in a few percent of cells needs many independent observations before it separates from sequencing error. Amplicon and hotspot panels are often planned at 500× to 1000×, which sounds extreme until you notice the target space is a few tens of kilobases. De novo assembly needs more than resequencing at the same genome size, because repeats must be spanned rather than merely aligned, and long-read assemblies are commonly planned in the 20× to 60× range depending on read-length distribution and error profile. RNA sequencing is usually planned in reads per sample rather than depth, because the effective target space is the expressed transcriptome and it changes between samples.

Paired-End, Single-End and Long-Read Coverage Math

The most common arithmetic slip on this page is double counting a paired-end run. A 2 × 150 run described as "400 million reads" may mean 400 million fragments (800 million reads, 120 Gb) or 400 million individual reads (60 Gb). The unit selector and the paired-end checkbox exist to make that choice explicit. Enter fragments and tick the box, or enter reads and leave it unticked; both routes give the same answer, and mixing them gives an answer that is wrong by a factor of two.

Paired-end data also buys something that does not appear in the depth number. Two reads from a fragment of known insert size give positional information across the unsequenced middle, which improves mapping in repetitive regions, supports structural variant detection and helps scaffold assemblies. Long reads change the calculation differently: mean depth is computed the same way, but with read lengths measured in kilobases the effective breadth for a given depth is better than the Poisson model predicts, because a single read spans a whole difficult region rather than starting inside it.

Cost, Storage and Compute Consequences of Deeper Coverage

Coverage decisions propagate well beyond the flow cell. Raw output scales linearly with depth, and so do FASTQ storage, transfer time, alignment wall-clock time and the cost of keeping BAM or CRAM files for the life of the project. A 30× human genome is roughly 120 Gb of raw output; at 60× it is 240 Gb, and a cohort of 500 samples turns that into a storage and compute programme rather than a sequencing order. Multiplexing several samples per flow cell, running a shallow pilot to measure the real duplicate rate and on-target fraction, and separating must-have depth from nice-to-have depth are the three levers that usually save the most money. Replicates matter too: if the design calls for technical or biological replicates, multiply both the depth target and the cost by the number of libraries before you present a budget.

Limitations and Assumptions Behind These Coverage Estimates

The model here is deliberately simple, and it is important to know where it stops being true. It assumes read start positions are uniform and independent across the reference, which is the assumption behind both the depth ratio and the Poisson breadth term. Real coverage is not uniform: GC-rich and GC-poor regions are systematically under-represented by PCR-based library preparation, homopolymers and low-complexity sequence produce local dropouts, and segmental duplications and repeat families absorb reads that cannot be uniquely placed. Capture panels add their own non-uniformity through variable probe efficiency, so a panel with a 30× mean can easily have exons at 5×.

The calculator also assumes a single fixed read length, whereas long-read platforms produce a broad length distribution, and it treats the usable fraction and duplicate rate as constants known in advance rather than quantities that depend on input DNA quality and library complexity. Duplicate rate in particular rises as you sequence deeper from a fixed number of input molecules, so a linear extrapolation from a shallow pilot will over-promise at depth. Finally, the reference size you enter defines what "coverage" means: entering a whole genome size for a capture experiment understates depth by two orders of magnitude, and entering a haploid size for a polyploid organism misstates it as well. Treat the output as a planning estimate to be checked against pilot metrics, not as a guarantee about any particular base.

DNA Sequencing Coverage Questions Answered

Sources Verified for This Coverage Calculator

The depth relationship C = N × L / G, the expected gap fraction e^−C and the Poisson threshold model used above were verified against the primary and review literature listed below. Depth recommendations by application are planning conventions summarised in the Sims review and vendor guidance, not fixed standards.

• Lander ES, Waterman MS. Genomic mapping by fingerprinting random clones: a mathematical analysis. Genomics, 1988;2(3):231–239. doi:10.1016/0888-7543(88)90007-9 — the origin of the coverage and gap equations used here.
• Sims D, Sudbery I, Ilott NE, Heger A, Ponting CP. Sequencing depth and coverage: key considerations in genomic analyses. Nature Reviews Genetics, 2014;15:121–132. doi:10.1038/nrg3642 — depth and breadth definitions, application-specific depth guidance and sources of coverage non-uniformity.
• Illumina. Coverage depth: sequencing coverage calculation and recommendations. illumina.com coverage guidance — vendor read-count and output figures used to sanity-check the worked example.