- Sequencing depth, coverage, read length, and error profiles are fundamental concepts in DNA and RNA sequencing. They influence the reliability of sequencing results, the detection of genetic variants, genome assembly quality, transcript quantification, and the interpretation of biological data. Although these terms are closely related, they describe different properties of sequencing experiments. Understanding their differences is essential when designing sequencing projects, selecting sequencing technologies, evaluating data quality, and interpreting bioinformatics results.
- Sequencing depth generally describes how many sequencing reads or read bases represent a particular position or region in a biological reference sequence. It is commonly expressed as the number of reads covering a nucleotide position or as an average depth across a genomic region. For example, if a particular nucleotide position is represented by 30 sequencing reads, its sequencing depth is 30×. A depth of 100× means that approximately 100 reads cover that position. In practice, depth varies across genomic positions because of differences in DNA fragmentation, library preparation, amplification, sequencing efficiency, mapping quality, and other technical factors.
- Average sequencing depth is calculated by dividing the total number of aligned sequencing bases by the size of the reference region being analysed. For a whole-genome sequencing experiment, the reference genome size is generally used as the denominator. For targeted sequencing or whole-exome sequencing, the calculation may instead use the total size of the targeted regions. The basic formula is: Average sequencing depth = Total number of aligned bases / Reference region size
- For example, if an experiment generates 90 billion aligned bases and the reference genome is 3 billion bases in size, the average sequencing depth is 30×. This means that, averaged across the genome, each reference position is represented by approximately 30 aligned sequencing bases. However, this average does not mean that every position has exactly 30 reads. Some regions may have substantially higher depth, while others may have low coverage or no reads at all.
- Sequencing depth is important because multiple observations of the same genomic position can help distinguish genuine biological differences from sequencing errors. When a nucleotide is supported by several independent, high-quality reads, confidence in the observed base generally increases. This is particularly important in variant detection, where sequencing data are examined to identify differences between an individual or sample and a reference genome. Higher depth can improve the detection of low-frequency variants, mosaic mutations, tumour subclones, and variants present in mixed microbial populations. Nevertheless, increasing depth does not automatically guarantee accurate results. Systematic sequencing errors, mapping errors, contamination, PCR duplicates, and biased library preparation can affect many reads simultaneously. Independent evidence and appropriate quality filters remain essential.
- Coverage is a related but distinct concept. It describes how much of a reference genome, target region, or other sequence space is represented by sequencing reads. Coverage can refer to the proportion of genomic positions receiving at least one read, or to the proportion receiving a specified minimum depth. For example, if 95% of the targeted genomic positions have at least 10 reads, the experiment has 95% coverage at 10× or greater. Coverage is therefore often reported together with a depth threshold.
- The distinction between sequencing depth and coverage is important. Depth describes the number of reads or bases representing a position, whereas coverage describes the proportion of positions represented at a specified depth. An experiment can have a high average depth but incomplete coverage if a small number of regions receive exceptionally high numbers of reads while other regions remain poorly represented. Conversely, an experiment may cover nearly all genomic positions but have relatively low depth at many positions. Both measurements are necessary to evaluate sequencing completeness and reliability.
- Coverage is commonly expressed as a percentage. For example, if a reference region contains 1,000,000 nucleotide positions and 950,000 positions have at least one aligned read, the coverage at 1× is 95%. If only 800,000 positions have at least 20 reads, coverage at 20× is 80%. These measurements provide more information than average depth alone because they reveal how uniformly sequencing data are distributed across the analysed region.
- The relationship between sequencing depth and genome coverage can also be described using probability models. Under an idealized random sequencing model, the probability that a position receives no reads decreases as sequencing depth increases. If reads are distributed uniformly and independently, the expected fraction of positions covered at least once can be approximated using the Poisson distribution: Coverage fraction = 1 – e^(-C)
- Here, C represents the average sequencing depth and e is the mathematical constant approximately equal to 2.71828. At an average depth of 1×, the expected fraction of positions receiving at least one read is approximately 63.2%. At 5×, it is approximately 99.3%. These values assume random and uniform sampling, which is rarely achieved perfectly in real sequencing experiments. GC bias, repetitive regions, uneven amplification, DNA quality, and mapping difficulties can produce substantial deviations from the theoretical expectation.
- Sequencing coverage is also affected by read length and the total number of reads generated. Read length refers to the number of nucleotide bases determined in an individual sequencing read. For example, a 150-base read contains 150 nucleotide positions. Short-read sequencing platforms commonly generate reads ranging from approximately 50 to 300 bases, depending on the platform and sequencing configuration. Long-read technologies, including Pacific Biosciences (PacBio) and Oxford Nanopore Technologies (ONT), can generate reads thousands or tens of thousands of bases long, with some reads extending beyond 100 kilobases.
- The total number of sequenced bases is determined by the number of reads and their lengths. For single-end sequencing, the total number of bases is approximately the number of reads multiplied by the average read length. For paired-end sequencing, both reads from each DNA fragment contribute to the total number of sequenced bases. The following formula illustrates this relationship: Total sequenced bases = Number of read pairs × 2 × Read length
- For example, if an Illumina experiment generates 100 million paired-end read pairs, with each read containing 150 bases, the total number of sequenced bases is: 100,000,000 × 2 × 150 = 30,000,000,000 bases
- This corresponds to 30 billion sequenced bases, or approximately 30 gigabases (Gb). If these reads are aligned to a reference genome of 3 gigabases, the theoretical average depth before accounting for filtering and alignment losses is approximately 10×. The actual usable depth may be lower because some reads may fail quality filtering, map ambiguously, represent PCR duplicates, or originate from contamination.
- Read length affects the types of biological questions that can be addressed using sequencing data. Short reads are generally suitable for detecting many single-nucleotide variants, small insertions and deletions, gene expression patterns, and microbial community composition. They often provide high base-level accuracy and are relatively economical to generate at large scale. However, short reads may be difficult to map uniquely in repetitive regions, segmental duplications, and complex genomic regions. They may also fail to span large insertions, structural rearrangements, and repetitive sequences.
- Long reads can span larger genomic regions and provide information about structural variation, repetitive DNA, haplotype structure, complex genomic rearrangements, and full-length transcript isoforms. They are particularly valuable for de novo genome assembly because long reads can connect genomic regions that cannot be reliably assembled using short reads alone. PacBio HiFi sequencing combines long reads with high consensus accuracy, while Oxford Nanopore sequencing provides a broad range of read lengths and can support direct sequencing of native DNA or RNA molecules. The accuracy and error characteristics of these technologies differ, so the choice of platform should depend on the biological objective and analytical requirements.
- Read length should not be confused with sequencing depth. A sequencing experiment can generate many short reads or fewer long reads while achieving a similar total number of sequenced bases. For example, 30 gigabases of short-read data and 30 gigabases of long-read data may provide similar theoretical average depth for the same reference genome, but they do not provide equivalent information. Long reads can span complex regions and reveal long-range sequence relationships, whereas short reads may provide greater numbers of independent observations and high base-level accuracy at a lower cost per base, depending on the platform and project.
- Sequencing depth is also influenced by the distinction between raw depth and effective depth. Raw depth is calculated using the total number of sequenced bases or aligned reads before all quality and duplication filters have been applied. Effective or usable depth refers to the sequencing information remaining after relevant filtering steps. These steps may include removal of low-quality reads, adapter trimming, exclusion of contaminant sequences, duplicate marking or removal, and filtering of poorly mapped reads. In variant analysis, effective depth may also be restricted to reads with sufficiently high mapping quality and base quality at the position being examined.
- PCR duplicates are an important consideration when calculating usable sequencing depth. During library preparation, DNA fragments may be amplified through polymerase chain reaction (PCR). Multiple reads can therefore originate from the same original DNA molecule. Counting all these reads as independent observations may overestimate the amount of independent molecular evidence supporting a variant. Duplicate marking tools such as Picard MarkDuplicates and SAMtools markdup help identify duplicate reads in suitable sequencing workflows. Unique molecular identifiers (UMIs) can provide additional information about the original molecules and support molecular deduplication or consensus generation. However, duplicate handling must be adapted to the sequencing protocol because repeated observations can have different meanings in different experimental designs.
- Sequencing depth requirements vary according to the biological application. Whole-genome sequencing for germline variant discovery commonly uses moderate or high average depth, depending on the organism, sample quality, sequencing platform, variant classes, and required sensitivity. Human germline whole-genome sequencing is often performed at approximately 30× average depth in many research and clinical workflows, although different applications may require different levels. Higher depth may be necessary for detecting low-frequency variants, mosaicism, or variants in heterogeneous samples. Tumour sequencing may require substantially greater depth, particularly when the tumour contains a low proportion of cancer cells or when rare subclonal mutations are important.
- Whole-exome sequencing generally requires higher average depth within targeted coding regions than whole-genome sequencing because only a relatively small portion of the genome is captured. Research and clinical workflows often target average exome depths around 50–100× or higher, depending on the intended application and laboratory requirements. However, the average depth should not be interpreted as proof that all clinically relevant exons are adequately covered. Capture efficiency varies across regions, and some exons may remain poorly covered despite a high overall average.
- Targeted gene panel sequencing often uses higher depths than whole-genome or whole-exome sequencing because the analysis is restricted to a smaller set of genomic regions. Depths of several hundred reads or more may be used when detecting low-frequency variants, including variants in cancer samples. The appropriate threshold depends on assay validation, the proportion of tumour cells, sequencing error rates, molecular barcoding, and the required variant allele frequency sensitivity. High nominal depth without sufficient independent molecules does not necessarily provide reliable evidence for very rare variants.
- RNA sequencing has different depth requirements because its primary objective is often the measurement of transcript abundance rather than the uniform representation of genomic positions. The required number of reads depends on the organism, transcriptome complexity, expression levels, library preparation method, and research question. Gene-level expression analysis may require fewer reads than transcript-level isoform analysis, rare transcript detection, allele-specific expression, or the study of low-abundance RNA species. Single-cell RNA sequencing requires additional considerations because reads are distributed among individual cells, and many transcripts are detected only in a subset of cells.
- Metagenomic sequencing also requires application-specific depth planning. In microbial community analysis, sequencing depth determines how many organisms and genes can be detected, especially when rare taxa are present. A low-abundance organism may be missed in a shallow sequencing dataset even when it is biologically important. However, sequencing depth alone cannot correct biases introduced during sample collection, DNA extraction, amplification, or library preparation. Reference database completeness and taxonomic classification methods also influence the detection of microbial species.
- Sequencing error profiles describe the types, frequencies, and distributions of errors produced during sequencing. Errors may involve substitution of one nucleotide for another, insertion of additional bases, or deletion of bases from the observed sequence. These errors can arise from the chemistry of the sequencing platform, signal detection, basecalling algorithms, sample preparation, amplification, and other technical processes. Different sequencing technologies have characteristic error patterns, and understanding them is essential for selecting appropriate quality-control and analytical methods.
- Substitution errors occur when a nucleotide is incorrectly identified as another nucleotide. For example, a true adenine may be reported as guanine. Substitutions are particularly relevant to single-nucleotide variant analysis because sequencing errors can resemble genuine genetic variants. Their frequency and distribution depend on the sequencing technology, basecalling process, sequence context, and quality of the data. In short-read sequencing, substitution errors are commonly evaluated using base quality scores and platform-specific quality-control metrics.
- Insertion errors occur when additional bases are reported that are not present in the original sequence. Deletion errors occur when bases present in the original molecule are omitted from the sequence output. Insertions and deletions are collectively called indel errors. They can be particularly problematic in homopolymer regions, where the same nucleotide is repeated several times. Some sequencing technologies have historically experienced difficulties determining the exact length of long homopolymers. Modern chemistry and basecalling improvements have reduced several error types, but homopolymer-associated errors remain important to assess in relevant datasets.
- Sequencing errors may be random or systematic. Random errors occur unpredictably and may differ between individual reads. When independent observations are combined, random errors can sometimes be reduced through consensus methods or repeated sequencing. Systematic errors occur more consistently in particular sequence contexts, genomic regions, or technical conditions. They are more difficult to eliminate through increased sequencing depth because the same error may appear repeatedly. Examples include errors associated with particular sequence motifs, GC-rich or GC-poor regions, homopolymers, PCR amplification bias, and mapping ambiguity.
- The Illumina sequencing platform uses sequencing by synthesis with fluorescently labelled reversible terminator nucleotides. Its short reads generally have high base-level accuracy, although error rates can increase toward the ends of reads, in low-quality cycles, or in difficult sequence contexts. Errors may include substitutions, incorrect base calls, and quality deterioration. Quality profiles are therefore commonly examined across sequencing cycles to identify declining read quality or unusual patterns.
- PacBio sequencing uses single-molecule real-time sequencing. Raw subreads may contain errors, but circular consensus sequencing (CCS) repeatedly reads the same circularized DNA insert to generate highly accurate HiFi reads. Consensus generation reduces random errors by combining multiple observations of the same molecule. The final accuracy depends on the number and quality of passes, insert length, library preparation, and sequencing conditions. PacBio HiFi reads are particularly useful for applications requiring both long read lengths and high sequence accuracy.
- Oxford Nanopore sequencing identifies nucleotides by measuring changes in ionic current as DNA or RNA molecules pass through nanopores. The electrical signals are interpreted by basecalling algorithms to produce nucleotide sequences. Error profiles depend on pore chemistry, sequencing conditions, basecalling models, read length, sequence context, and the molecule type being sequenced. Insertions, deletions, and substitutions may occur, and homopolymer regions can present particular challenges. Improvements in pore technology, basecalling algorithms, and duplex sequencing have increased accuracy, but error profiles should still be evaluated for each dataset and application.
- Sequencing quality is commonly represented using Phred quality scores. A Phred score expresses the estimated probability that a base call is incorrect. The relationship is: Q = -10 × log10(P)
- Here, Q is the Phred quality score and P is the probability that the base call is incorrect. A Q20 score corresponds to an estimated error probability of 1%, while Q30 corresponds to an error probability of 0.1%. A Q40 score corresponds to an error probability of 0.01%. Higher Phred scores therefore indicate greater confidence in the base call. These scores represent estimated probabilities and depend on the accuracy of the platform’s quality calibration.
- Sequencing quality is also evaluated using read-level and run-level metrics. Important measurements include the number of reads, total bases, read length distribution, percentage of bases above selected quality thresholds, GC-content distribution, adapter contamination, ambiguous nucleotide frequency, duplication rate, and sequence complexity. For paired-end sequencing, the quality of forward and reverse reads may differ. For long-read sequencing, read length distribution, quality distribution, read yield, and platform-specific signal or basecalling metrics are particularly important.
- Several bioinformatics tools are widely used to assess sequencing quality. FastQC provides an overview of read quality, sequence composition, GC content, adapter contamination, duplication, and other quality indicators. MultiQC combines results from multiple quality-control tools into a consolidated report, making it useful for large sequencing projects. fastp performs quality filtering, adapter trimming, and other preprocessing operations for short-read data. For long-read data, NanoPlot and NanoStat can summarize read length and quality distributions, while platform-specific tools provide additional information about sequencing performance. The selection of tools depends on the sequencing technology and data type.
- The following command illustrates basic quality assessment of paired-end FASTQ files using FastQC:
fastqc sample_R1.fastq.gz sample_R2.fastq.gz -o fastqc_results/
- A MultiQC report can then be generated from the quality-control outputs:
multiqc fastqc_results/ -o multiqc_report/
- For short-read data, adapter trimming and quality filtering may be performed using fastp:
fastp \
-i sample_R1.fastq.gz \
-I sample_R2.fastq.gz \
-o sample_R1_clean.fastq.gz \
-O sample_R2_clean.fastq.gz \
-h fastp_report.html \
-j fastp_report.json
- The resulting reports should be examined before proceeding to alignment or other downstream analyses. Quality filtering should not be applied automatically without considering the biological purpose. Aggressive trimming can remove useful sequence information, reduce read length, and affect mapping or variant detection. Quality thresholds should therefore be selected according to the sequencing platform, experimental design, and downstream application.
- Sequencing depth and coverage can be evaluated after alignment using tools such as SAMtools, mosdepth, Qualimap, and Picard. SAMtools provides commands for calculating depth and alignment statistics. mosdepth efficiently calculates coverage and depth distributions across large genomic datasets. Picard provides additional alignment and sequencing metrics, while Qualimap offers quality-control summaries for different types of sequencing data.
- The following example calculates per-position sequencing depth using SAMtools:
samtools depth -a sample.bam > depth.txt
- The -a option includes positions with zero coverage in the output. The resulting file contains genomic coordinates and depth values. Average depth can be calculated from these values using a simple Python script:
total_depth = 0
position_count = 0
with open(“depth.txt”) as file:
for line in file:
chromosome, position, depth = line.strip().split(“\t”)
total_depth += int(depth)
position_count += 1
if position_count > 0:
average_depth = total_depth / position_count
print(“Average depth:”, average_depth)
else:
print(“No positions found.”)
For larger projects, mosdepth can generate coverage summaries and depth distributions more efficiently:
mosdepth sample sample.bam
- Coverage at a specified threshold can be assessed using the generated summary and region-level outputs, with the relevant reference intervals supplied when targeted coverage is required. Researchers should ensure that the denominator corresponds to the intended reference genome or target regions and that the treatment of unmapped reads, secondary alignments, supplementary alignments, duplicates, and low-quality mappings is appropriate for the analysis.
- Sequencing depth and error profiles are particularly important in variant calling. A variant caller evaluates the reads supporting alternative nucleotides or indels at a genomic position. The number of reads covering the position is commonly called read depth or DP. The number of reads supporting a particular alternative allele may be reported as alternative allele depth or AD. Variant allele frequency (VAF) is calculated by dividing the number of reads supporting the alternative allele by the total number of relevant reads at that position. For example, if 20 out of 100 reads support an alternative allele, the observed VAF is 20%. In diploid germline data, heterozygous variants often have VAF values near 50%, while homozygous alternative variants may have values near 100%, although biological and technical factors can cause deviations.
- In cancer sequencing, VAF can be affected by tumour purity, subclonal structure, copy-number changes, loss of heterozygosity, and normal-cell contamination. A mutation present in only a small fraction of tumour cells may have a low VAF and can be missed if sequencing depth is insufficient. Similarly, low-frequency variants in mosaic samples or mixed microbial populations may require higher depth and carefully validated error suppression. Unique molecular identifiers, duplex sequencing, and consensus-based methods can improve the reliability of rare-variant detection by distinguishing independent original molecules from repeated sequencing observations.
- The relationship between depth and variant detection can be illustrated using a simplified sampling model. If a variant is present at a true allele frequency of 5%, the probability of observing no alternative reads decreases as sequencing depth increases. However, detecting at least one alternative read is not sufficient for reliable variant identification. A practical variant caller requires evidence thresholds based on read quality, mapping quality, strand balance, allele fraction, local sequence context, and other statistical criteria. The assay’s validated limit of detection is therefore more meaningful than depth alone.
- Sequencing depth and read length also influence de novo genome assembly. Genome assembly algorithms reconstruct longer genomic sequences from overlapping reads or other sequence relationships. Short-read assembly can be limited by repetitive regions, duplicated genes, and complex structural variation because reads may not span the repeated sequences. Long reads can bridge these regions and improve contiguity, although their usefulness depends on read accuracy, length distribution, coverage, and the complexity of the genome. Assembly quality is commonly evaluated using metrics such as contig N50, total assembly length, number of contigs, completeness, and misassembly estimates. Tools such as QUAST, BUSCO, and CheckM2 are used for different aspects of assembly assessment, depending on whether the target is a eukaryotic genome, microbial genome, or metagenome-derived genome.
- Coverage uniformity is another important property of sequencing experiments. Ideally, sequencing reads are distributed relatively evenly across the reference or targeted regions. In practice, coverage can vary because of GC content, sequence complexity, fragment size, PCR amplification, hybridization capture efficiency, and mapping difficulty. GC-rich and GC-poor regions may be underrepresented in some workflows. Repetitive sequences may be difficult to map uniquely, and regions with structural variation may align poorly to a single linear reference genome. In targeted sequencing, capture probe design and hybridization efficiency can cause substantial differences in coverage between exons or genes.
- Coverage uniformity is often evaluated using the proportion of target bases reaching defined depth thresholds, such as 10×, 20×, 30×, or higher. Some workflows also report the coefficient of variation of coverage, which measures relative variation in depth across regions. A lower coefficient of variation generally indicates more uniform coverage, although its interpretation depends on the assay and the regions included. For clinical sequencing, laboratories may define minimum coverage requirements for reporting variants and identify regions that require supplementary testing when coverage is insufficient.
- The choice of sequencing depth, read length, and technology should be guided by the research objective rather than by a universal recommendation. For germline variant analysis, sufficient depth and uniform coverage are important for reliable genotype calls. For rare-variant detection, the required depth depends on the expected allele frequency and the assay’s error characteristics. For transcriptomics, read allocation should account for transcript abundance and the level of expression detail required. For metagenomics, depth should be sufficient to detect organisms and functional genes of interest. For genome assembly, read length, coverage, read accuracy, and repeat structure must be considered together.
- Sequencing experiments should also account for biological replication. Increasing sequencing depth within a single sample cannot replace the need for independent biological replicates when the research question involves comparing groups or estimating biological variation. For example, in differential gene expression analysis, adding more reads to one sample may improve measurement precision, but it does not provide the same statistical information as sequencing additional independent biological samples. Experimental design should therefore balance sequencing depth, sample number, read length, and sequencing cost.
- For reproducible bioinformatics analysis, researchers should document the sequencing platform, chemistry or kit where relevant, read configuration, read length, total read count, total bases, average depth, coverage thresholds, quality scores, error characteristics, and filtering procedures. They should also report the reference genome or target design, alignment software and version, duplicate-handling strategy, and criteria used to exclude low-quality reads. For long-read sequencing, basecaller version and model settings may be particularly important because changes in basecalling can affect accuracy and downstream results.
- Sequencing depth, coverage, read length, and error profiles are interconnected but measure different aspects of sequencing data. Depth describes how many reads or bases represent a position, coverage describes how much of the reference region is represented at a specified depth, read length describes the number of bases in each read, and error profiles describe the types and distributions of incorrect base calls. Together, these properties determine the amount, completeness, and reliability of sequencing information available for bioinformatics analysis. Their careful evaluation enables researchers to select appropriate sequencing strategies, detect variants more confidently, assemble genomes more effectively, quantify biological molecules, and produce reliable and reproducible scientific results.