Sequencing Coverage: Definition, Calculation, and Best Practices

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-11

Introduction to Sequencing Coverage

Sequencing coverage—also termed sequencing depth—is the average number of times a given nucleotide within a sequenced library is represented by a sequencing read. If a genome of size G base pairs is sequenced and produces N reads of average length L, the coverage C is expressed as:

C = (N × L) / G

A coverage value of 30× means that, on average, each base in the target region has been read 30 independent times. Coverage is the single most important quantitative parameter in next-generation sequencing (NGS) because it directly determines the statistical confidence of every downstream inference, from variant calling to transcript quantification.

Definition of Coverage

Coverage is a measure of redundancy. In a sequencing experiment, the instrument does not read the genome as a continuous string; it produces millions of short, independent reads that are then aligned to a reference or assembled de novo. The depth of coverage at a specific genomic position is the count of reads whose aligned intervals include that position. Genome-wide coverage is the mean of these per-position counts across all mappable positions.

The term "coverage" is also used in a second sense: the fraction of the genome that is covered by at least one read. To avoid ambiguity, the field now distinguishes between depth (the redundancy measure) and breadth (the fraction of the genome represented). Both are essential, but they answer different questions. Depth determines the confidence of a base call; breadth determines whether a locus is represented at all. A sequencing run can achieve high depth in a small region while leaving most of the genome uncovered, or it can achieve broad but shallow coverage across the entire genome.

Depth vs. Breadth

Depth and breadth are orthogonal metrics. Depth is a per-position count; breadth is a genomic proportion. For a diploid human genome of approximately 3.1 Gb, 30× depth with 95% breadth means that 95% of the genome has at least one read, and the average depth across those covered positions is roughly 31.6× (30 / 0.95). The remaining 5% of the genome—typically centromeres, telomeres, and highly repetitive regions—is refractory to short-read alignment and remains uncovered regardless of total output.

Breadth matters most in applications where the presence or absence of a sequence is the biological question, such as detecting structural variants, identifying microbial taxa in metagenomic samples, or confirming the absence of a pathogen. Depth matters most where the precision of a base call is critical, such as identifying a heterozygous single-nucleotide variant (SNV) or a low-frequency somatic mutation.

How Sequencing Coverage Is Calculated

The Basic Formula

The theoretical coverage of a sequencing run is calculated from three parameters: the number of reads produced (N), the read length (L), and the size of the target genome (G). For a whole-genome sequencing (WGS) experiment on a human sample with 300 million paired-end reads of 150 bp each, the calculation is:

C = (300 × 10⁶ × 150) / (3.1 × 10⁹) ≈ 14.5×

This is the theoretical or raw coverage. It assumes that every read is mappable, that every read is unique, and that the genome is uniformly accessible. None of these assumptions holds in practice.

Factors Affecting Actual Coverage

The actual coverage—the mean depth observed after alignment—is always lower than the theoretical value. Several factors contribute to this discrepancy:

1. Read filtering: During quality control, reads with low Phred scores, adapter contamination, or failed base calls are removed. This typically discards 1–5% of raw reads.

1. Mapping rate: Not all reads align to the reference. Reads originating from repetitive regions, structural variant breakpoints, or contaminating DNA (e.g., bacterial DNA in a human sample) fail to map. Mapping rates for human WGS typically range from 95–99%.

1. PCR duplicates: During library preparation, fragments are amplified by PCR. After sequencing, reads derived from the same original fragment are identified as duplicates (identical outer coordinates and orientation) and removed to avoid overcounting. Duplicate rates vary from 5% in high-quality libraries to over 30% in low-input samples. The Library Prep in Sequencing step is the primary determinant of duplicate rates; lower input amounts require more PCR cycles and produce more duplicates.

1. GC bias: The PCR amplification step preferentially amplifies fragments with moderate GC content (40–60%). Regions with extreme GC content are under-represented in the final library, leading to lower coverage in those regions.

1. Mappability: Regions with high sequence similarity to other genomic locations (e.g., segmental duplications, pseudogenes) produce reads that align ambiguously. Most aligners discard these reads or assign them a low mapping quality, effectively reducing coverage in those regions.

The ratio of actual to theoretical coverage is termed the efficiency or yield of the run. For a well-optimized Illumina WGS run, efficiency is typically 80–90%.

The Role of Coverage in Variant Calling

Minimum Depth for Reliable Calls

Variant calling is a statistical exercise. At each genomic position, the caller observes a set of bases from aligned reads and must decide whether the observed variation from the reference is a true biological variant or an artifact of sequencing error. The probability of a false call decreases with increasing depth.

For a diploid sample, a heterozygous SNV should be supported by approximately 50% of the reads covering that position. If the sequencing error rate is 0.1% (a Phred score of 30), then at 10× depth, the expected number of error-derived alternate alleles is 0.01 (10 × 0.001). The probability of observing at least one alternate allele purely from error is roughly 1%. At 30× depth, the expected number of error-derived alternate alleles is 0.03, but the probability of observing enough alternate alleles to mimic a heterozygote (e.g., 10 of 30 reads) is astronomically small. This is why 30× is the standard for human germline WGS: it provides overwhelming statistical separation between true heterozygotes and sequencing artifacts.

For somatic variant calling in cancer samples, higher depth is required because the variant allele fraction (VAF) can be low. A tumor sample with 40% purity has a maximum VAF of 20% for a heterozygous somatic mutation. To detect a 5% VAF mutation with confidence, the depth must be high enough that the mutation is represented by multiple reads. At 100× depth, a 5% VAF mutation is expected to appear in 5 reads—sufficient for detection but marginal for confident calling. Most clinical somatic panels use 500–1000× depth.

Coverage and Allele Frequency

The relationship between coverage and the minimum detectable allele frequency follows a binomial distribution. If the true allele frequency is f and the depth is d, the probability of observing k alternate reads is:

P(k) = C(d, k) × fᵏ × (1 − f)^(d−k)

To detect a variant with confidence, the observed alternate read count must exceed the expected count from sequencing errors. The limit of detection (LOD) for a variant at a given depth is approximately:

LOD ≈ 3 × e / d

where e is the sequencing error rate. At 30× depth with a 0.1% error rate, the LOD is approximately 0.01%—far below the 50% expected for a germline heterozygote. At 1000× depth (typical of targeted panels), the LOD drops to 0.0003%, enabling detection of rare subclonal mutations.

Coverage Requirements for Different Applications

The optimal coverage depth depends on the biological question, the complexity of the genome, and the error rate of the sequencing platform. The table below summarizes typical coverage recommendations.

Whole-Genome Sequencing

For human germline WGS, 30× is the consensus standard. This depth provides >99% sensitivity for SNV detection and >95% sensitivity for small indels. The 1000 Genomes Project and the Genome in a Bottle Consortium both used 30× as their reference standard. Higher depth (60×) is sometimes used for clinical diagnostic applications where the cost of a missed variant is high, but the marginal benefit beyond 30× is modest for germline variants.

Exome Sequencing

Exome sequencing requires higher mean depth than WGS because the capture process is inherently non-uniform. The exome is enriched by hybridization to biotinylated probes, and the efficiency of capture varies across targets. A typical exome run produces a mean depth of 100×, but the distribution is broad: some regions exceed 300× while others fall below 20×. The standard quality metric for exome sequencing is the fraction of target bases covered at ≥20×, which should exceed 80% for clinical use.

RNA-Seq

RNA-seq coverage is typically expressed as the number of reads per sample rather than depth per base, because the "genome" being sequenced is the transcriptome, which varies in composition across cell types. For gene expression quantification, 25–50 million 75-bp paired-end reads per sample is sufficient for detecting moderate-to-high abundance transcripts. For isoform-level analysis or detection of low-abundance transcripts, 100–200 million reads are recommended. The Prepare Sample for Sequencing step is critical for RNA-seq because ribosomal RNA depletion and poly-A selection dramatically affect the distribution of reads across transcripts.

Targeted Panels

Targeted sequencing panels—whether amplicon-based or hybrid-capture—are designed to interrogate a small set of genomic regions at very high depth. Amplicon panels (e.g., Ion AmpliSeq, Illumina TruSeq Amplicon) use multiplex PCR to amplify specific loci. Because the target is small (typically 1–500 kb), the same sequencing output that yields 30× for a genome yields 1000× for a panel. This high depth enables detection of somatic mutations at VAFs as low as 1–5%, which is essential for liquid biopsy applications. The Sanger Sequencing Protocol remains the gold standard for validating variants identified by targeted panels, though it requires the variant VAF to exceed approximately 15% for reliable detection.

Methods to Measure and Assess Coverage

Common Tools

Several bioinformatics tools compute coverage metrics from aligned BAM files:

samtools depth: This command calculates per-position depth across the genome. It outputs three columns: chromosome, position, and depth. The output can be piped to awk to compute mean, median, and percentiles. For example:

samtools depth -a sample.bam | awk '{sum+=$3; sumsq+=$3*$3} END {print "mean=" sum/NR, "stdev=" sqrt(sumsq/NR - (sum/NR)^2)}'

The -a flag forces output for all positions, including those with zero depth.

Picard CollectWgsMetrics: This tool computes coverage metrics specifically for WGS data. It reports mean depth, the fraction of bases covered at various thresholds (1×, 10×, 20×, 30×), and the uniformity of coverage. The tool requires a reference genome and a BAM file with mapped reads.

GATK DepthOfCoverage: This tool provides per-interval coverage statistics and is useful for exome or targeted panel data. It can output coverage histograms and per-sample summaries.

Qualimap: This tool provides a graphical summary of coverage, including a coverage histogram, GC-content bias plots, and per-chromosome coverage profiles.

Coverage Uniformity and GC Bias

Mean coverage is an incomplete descriptor of sequencing quality. Two samples can both have 30× mean coverage, but one may have uniform coverage across the genome while the other has extreme variation. The standard metrics for uniformity are:

• Coefficient of variation (CV): The ratio of the standard deviation to the mean. A CV below 0.5 indicates good uniformity; values above 1.0 indicate severe non-uniformity.
• Fold-80 penalty: The ratio of the mean depth to the depth below which 80% of the genome falls. A fold-80 penalty of 1.5 means that 80% of the genome is covered at a depth at least 1.5-fold below the mean. Lower values indicate better uniformity.
• Percent of bases at ≥10× and ≥20×: These thresholds are clinically relevant because they define the regions where variant calling is reliable.

GC bias is the primary driver of non-uniformity in short-read sequencing. During PCR amplification in library preparation, fragments with extreme GC content amplify less efficiently than those with moderate GC content. The result is a U-shaped relationship between GC content and coverage: regions with 40–60% GC are well covered, while regions with <20% or >80% GC are under-covered. This bias is particularly problematic for exome sequencing, where GC-rich first exons are common. Methods to mitigate GC bias include using PCR-free library preparation, optimizing PCR cycling conditions, and applying computational GC correction during variant calling.

Coverage and Genome Assembly

Lander-Waterman Model

In de novo genome assembly, coverage determines the contiguity and completeness of the assembled genome. The Lander-Waterman model, developed by Eric Lander and Michael Waterman in 1988, describes the relationship between coverage, read length, and assembly contiguity.

The model assumes that reads are randomly distributed across the genome. The expected number of gaps (regions not covered by any read) in the assembly is:

G × e^(−C)

where G is the genome size and C is the coverage. At 10× coverage, the expected fraction of the genome in gaps is e^(−10) ≈ 0.0045%, which is negligible. However, the model also predicts the expected number of contigs (contiguous assembled sequences). The number of contigs is approximately:

N × e^(−C)

where N is the number of reads. At low coverage, the genome is fragmented into many small contigs; at high coverage, the contigs merge into larger scaffolds.

Optimal Coverage for Assembly

The optimal coverage for assembly depends on the genome complexity and the read length. For bacterial genomes (5 Mb), 50–100× coverage with 150-bp reads is sufficient to produce a complete, circular chromosome. For eukaryotic genomes, the situation is more complex because of repetitive elements. A repeat that is longer than the read length cannot be resolved by short-read assembly alone, regardless of coverage. This is why the Nanopore vs Sanger Sequencing comparison is relevant: long-read platforms (Oxford Nanopore, PacBio) produce reads of 10–100 kb that span repetitive regions, enabling complete assembly of eukaryotic genomes with lower coverage (30–50×) than short-read assembly (100–200×).

The trade-off between depth and contiguity is not monotonic. Increasing coverage beyond a certain point yields diminishing returns because the additional reads are mostly duplicates of already-sequenced fragments. For short-read assembly, the optimal coverage is typically 50–100× for bacteria and 100–200× for eukaryotes. Beyond 200×, the computational cost of assembly increases without proportional improvement in contiguity.

Common Pitfalls in Interpreting Coverage

Low Coverage Regions

A common error is to treat mean coverage as if it applied uniformly across the genome. In practice, coverage varies by orders of magnitude between regions. Low-coverage regions are not random; they are concentrated in GC-rich or GC-poor regions, repetitive elements, and segmental duplications. Variants called in these regions have a higher false-positive rate because the few reads that do align may be of low quality or derived from paralogous sequences.

When interpreting a variant call, always check the depth at that specific position. A variant called at 5× depth in a 30× genome is not equivalent to a variant called at 30× depth. Most variant callers annotate the depth and allele balance for each call; these fields should be examined before downstream analysis.

Mapping Quality

Coverage counts reads, but not all reads are equally reliable. Mapping quality (MAPQ) is a Phred-scaled probability that a read is incorrectly placed. A read with MAPQ 0 is equally likely to map to multiple locations; a read with MAPQ 60 has a 1 in 10⁶ chance of being mis-mapped. Counting low-MAPQ reads in coverage inflates the apparent depth at repetitive regions and can lead to false variant calls.

Best practice is to filter reads by MAPQ (typically ≥20) before computing coverage or calling variants. Tools like samtools view -q 20 apply this filter. Additionally, base quality should be considered: bases with Phred quality below 20 should be excluded from coverage calculations.

Amplicon vs. Shotgun

Coverage has different meanings in amplicon and shotgun sequencing. In shotgun sequencing, reads are randomly distributed across the genome, and coverage follows a Poisson distribution. In amplicon sequencing, reads are concentrated at specific loci, and coverage within an amplicon is highly uniform but zero outside it. The mean coverage across the entire genome is meaningless for amplicon data; the relevant metric is the depth within each amplicon.

Furthermore, amplicon data are subject to allelic dropout: if a primer binding site contains a polymorphism, the primer may fail to anneal, and the allele is not amplified. This leads to false homozygosity calls. Amplicon-based approaches are therefore less reliable than hybrid-capture for clinical variant detection, despite their higher depth. The CHIP Sequencing protocol, by contrast, uses immunoprecipitation followed by shotgun sequencing, and its coverage requirements are governed by the same Poisson statistics as WGS.

Practical Summary: Choosing the Right Coverage

Decision Framework

Selecting the appropriate coverage depth requires balancing statistical power, cost, and the specific biological question. Use the following framework:

1. Define the variant classes of interest. Germline SNVs require 30×; somatic mutations at 5% VAF require 500×; structural variants require high breadth rather than high depth.

1. Consider the genome complexity. Repetitive genomes (plants, some parasites) require higher coverage for the same variant detection sensitivity because a fraction of reads will be unmappable.

1. Account for sample purity. Tumor samples with low purity require higher depth to detect somatic mutations at reduced VAF.

1. Check platform error rates. Platforms with higher error rates (e.g., Oxford Nanopore) require higher coverage to achieve the same base-level accuracy as lower-error platforms (e.g., Illumina).

1. Plan for downstream analyses. If the data will be used for multiple purposes (variant calling, copy number analysis, structural variant detection), choose the highest coverage required by any single analysis.

Cost-Benefit Considerations

The cost of sequencing scales linearly with coverage, but the benefit follows a diminishing-returns curve. For germline WGS, increasing from 10× to 30× substantially improves variant detection sensitivity; increasing from 30× to 60× provides only marginal improvement. For somatic panels, the relationship is different: increasing from 500× to 1000× doubles the sensitivity for rare mutations, which may be clinically significant.

A cost-effective strategy is to use a tiered approach: sequence at moderate depth (30×) for discovery, then validate candidate variants with targeted deep sequencing (500–1000×) or Sanger Sequencing Method. This approach minimizes cost while maintaining high sensitivity and specificity.

Frequently Asked Questions

What is sequencing coverage?

Sequencing coverage (depth) is the average number of times each nucleotide in a genome or target region is represented by a sequencing read. It is calculated as the total number of bases sequenced divided by the size of the target genome.

What is coverage in sequencing?

Coverage in sequencing refers to the redundancy of sequence data. A coverage of 30× means that, on average, each base in the genome has been sequenced 30 times. Coverage is distinct from breadth, which is the fraction of the genome represented by at least one read.

How do you calculate sequencing coverage?

Sequencing coverage is calculated using the formula: Coverage = (Number of reads × Read length) / Genome size. For example, 100 million 150-bp reads from a 3.1-Gb human genome yield a theoretical coverage of (100 × 10⁶ × 150) / (3.1 × 10⁹) ≈ 4.8×.

What is a good sequencing coverage for whole genome sequencing?

For human germline WGS, 30× is the standard. This depth provides >99% sensitivity for SNV detection. For clinical applications where sensitivity is critical, 60× may be used. For non-human genomes, the optimal coverage depends on genome size and complexity.

What does 30x coverage mean?

30× coverage means that each base in the genome has been sequenced, on average, 30 times. This depth provides sufficient statistical power to distinguish true heterozygous variants from sequencing errors in diploid genomes.

Why is high coverage important in sequencing?

High coverage increases the confidence of base calls by providing multiple independent observations of each nucleotide. It enables detection of low-frequency variants, reduces the impact of sequencing errors, and improves the accuracy of de novo assembly.

What is the difference between coverage and depth?

Coverage and depth are often used interchangeably, but a distinction exists: depth refers to the number of reads at a specific position, while coverage can refer to either depth or breadth (the fraction of the genome covered). In practice, "coverage" usually means depth.

How does coverage affect variant calling?

Coverage determines the statistical confidence of variant calls. Low coverage increases the probability of false negatives (missing true variants) and false positives (calling errors as variants). Higher coverage enables detection of variants at lower allele frequencies and improves the accuracy of genotype calls.

Key Takeaways

• Sequencing coverage (depth) is the average number of reads per base and is calculated as (read count × read length) / genome size.
• Depth and breadth are distinct metrics: depth measures redundancy, breadth measures the fraction of the genome represented.
• Actual coverage is always lower than theoretical coverage due to read filtering, mapping failures, PCR duplicates, and GC bias.
• Germline WGS requires 30× depth; somatic mutation detection requires 500–1000×; RNA-seq requires 25–200 million reads depending on the analysis.
• Coverage uniformity matters as much as mean depth; the fold-80 penalty and coefficient of variation are key metrics.
• In de novo assembly, the Lander-Waterman model predicts that coverage determines contiguity, but long-read platforms reduce the coverage needed for complex genomes.
• Always filter reads by mapping quality before computing coverage, and interpret depth at individual variant positions rather than relying solely on genome-wide means.

Further Reading

• Ekblom R, Smeds L, Ellegren H. Patterns of sequencing coverage bias revealed by ultra-deep sequencing of vertebrate mitochondria. BMC genomics. 2014. PubMed 24923674
• Porębski S, Stokowy T. eXNVerify: coverage analysis for long and short-read sequencing data in clinical context. F1000Research. 2022. PubMed 38173955
• Lamb HJ et al. Genomic prediction using low-coverage portable Nanopore sequencing. PloS one. 2021. PubMed 34910782
• Rozhoňová H et al. SECEDO: SNV-based subclone detection using ultra-low coverage single-cell DNA sequencing. Bioinformatics (Oxford, England). 2022. PubMed 35900151
• Sanghvi RV et al. Characterizing reduced coverage regions through comparison of exome and genome sequencing data across 10 centers. Genetics in medicine : official journal of the American College of Medical Genetics. 2018. PubMed 29144510
• Chen J et al. AKSmooth: enhancing low-coverage bisulfite sequencing data via kernel-based smoothing. Journal of bioinformatics and computational biology. 2014. PubMed 25553811

Related Clinical & Scientific Guides

• MAPK Pathway: Mechanism, Function, and Clinical Relevance
• Mammalian Cell Culture Bioreactors: A Practical Guide
• Nucleotide Formation: Biosynthesis and Assembly of DNA/RNA Building Blocks