Whole-genome sequencing (WGS) is the determination of the DNA sequence across an organism’s genome, rather than only selected genes or genomic regions. It combines laboratory methods that read DNA molecules with computational methods that reconstruct sequences and identify differences among samples. Depending on the organism and study design, it may encompass nuclear chromosomes, organelle genomes, and additional DNA molecules such as plasmids. “Whole” describes the intended scope of analysis; it does not guarantee that every base, chromosome copy, or difficult repetitive region has been resolved.
WGS provides a common experimental foundation for investigating inherited disease, cancer, microbial transmission, population history, and biological diversity. Its importance lies partly in its breadth: the same dataset can reveal changes in protein-coding sequences, regulatory regions, chromosome structure, and other parts of the genome. The information recovered nevertheless depends on sample quality, sequencing technology, computational analysis, and the biological questions being asked. Producing sequence data, establishing an accurate genome sequence, and interpreting its biological significance are related but distinct tasks.
Scope and relationship to other genomic methods
A genome contains more than a catalogue of genes. In humans, a haploid nuclear genome contains approximately 3.1 billion base pairs distributed among a set of chromosomes, while protein-coding exons account for roughly 1–2% of the sequence. Much of the remainder consists of introns, regulatory sequences, repetitive DNA, and regions with incompletely characterized functions. WGS attempts to sample this broad landscape without first restricting sequencing to a predefined collection of targets. Its effective scope therefore differs from that of a gene panel, which examines selected genes, or whole-exome sequencing, which enriches primarily for protein-coding exons.
The distinction is experimental rather than absolute. Most WGS workflows sequence many DNA fragments and infer their genomic positions afterward. Exome sequencing and many gene-panel assays instead use capture probes or amplification to concentrate on specified regions before sequencing. Targeted assays can devote more reads to those regions for a given sequencing output, which is useful when detecting uncommon variants in mixed samples. WGS generally provides more uniform coverage outside capture-dependent regions and can support a broader range of analyses, including some that were not anticipated when the sample was collected.
Two major objectives are resequencing and de novo sequencing. Resequencing compares an individual sample with an existing reference genome to identify sequence differences. De novo sequencing seeks to reconstruct the sample’s sequence without requiring an existing reference to determine the arrangement of its bases. The resulting genome assembly may consist of many disconnected sequences or, with suitable data, nearly complete chromosomes. A dataset suitable for detecting common single-base variants is not necessarily sufficient to assemble every chromosome or reconstruct both parental copies separately.
The definition of the biological genome also matters. A human sample usually contains nuclear DNA and numerous copies of the mitochondrial genome within mitochondria. Plant samples can additionally contain chloroplast DNA. Bacterial genomes may include a main chromosome and one or more plasmids. Extraction methods and library preparation can preferentially retain or lose particular molecules, so nominal whole-genome sequencing does not ensure equal representation of all genomic components.
WGS is distinct from measurements of gene expression, which examine which genes are being used in a particular tissue or condition. Conventional DNA sequencing does not directly measure RNA abundance, protein concentration, or most biochemical activities. Some sequencing platforms can detect native DNA modifications, but ordinary WGS should not be treated as a comprehensive assay of epigenetic state. Sequence provides a relatively stable molecular framework; cellular activity also depends on developmental history, environment, and regulatory processes.
Historical development and reference genomes
The development of WGS followed advances in DNA sequencing and in the computation needed to organize large numbers of short sequence observations. Frederick Sanger and colleagues introduced the chain-termination method in 1977, the same year that the complete sequence of the bacteriophage phi X174 genome was published. At 5,386 nucleotides, that viral genome demonstrated that an entire genetic system could be represented as a continuous sequence. Sanger sequencing subsequently became the principal technology for many early genome projects, although its relatively limited throughput required extensive laboratory organization.
Larger genomes were commonly approached through fragmentation and reconstruction. In shotgun sequencing, DNA is broken into fragments, many fragments are sequenced, and overlaps are used to infer their original arrangement. In 1995, the approximately 1.83-million-base-pair genome of Haemophilus influenzae became the first published complete genome of a free-living organism. The project demonstrated the effectiveness of whole-genome shotgun sequencing at bacterial scale. The budding yeast Saccharomyces cerevisiae genome followed in 1996, providing a landmark reference for a eukaryotic organism.
The Human Genome Project, formally launched in 1990, combined large-scale sequencing with genetic and physical mapping. Public-project and Celera draft human genome analyses were published in 2001, and completion of the Human Genome Project was announced in 2003. “Completion” referred to the project’s defined targets and standards rather than an uninterrupted sequence through every chromosomal region. Highly repetitive regions, including substantial portions of centromeres and the short arms of acrocentric chromosomes, remained difficult to reconstruct.
Massively parallel sequencing transformed the scale of genome research during the mid-2000s. Instead of processing relatively small numbers of sequencing reactions at a time, instruments read millions of fragments in parallel. This development enabled individual human genomes, large population collections, and extensive microbial surveillance to be studied with substantially greater throughput. Projects such as the 1000 Genomes Project, launched in 2008, used large-scale sequencing to characterize human genetic variation across multiple populations and established widely used resources for variant interpretation.
Long-read technologies subsequently addressed regions that short reads could not reliably reconstruct. In 2022, the Telomere-to-Telomere Consortium published a substantially complete human genome assembly, T2T-CHM13, that resolved many sequences absent from earlier references; this assembly did not include a Y chromosome. A complete human Y chromosome sequence was published in 2023. Also in 2023, the Human Pangenome Reference Consortium published a draft human pangenome reference incorporating multiple diverse genome assemblies. These developments changed the reference from a single convenient coordinate framework toward a representation that can accommodate alternative sequences and chromosome structures.
Samples, DNA preparation, and experimental design
WGS begins with a biological sample, not with a sequencing instrument. Blood, saliva, cultured cells, fresh or frozen tissue, microbial colonies, and environmental material can all provide DNA, but they differ in cellular composition, contamination risk, and molecular preservation. A blood sample typically supplies DNA from nucleated blood cells rather than mature erythrocytes, which lack nuclei in humans. A tumor biopsy may contain malignant cells, immune cells, connective tissue, and necrotic material. Consequently, the sequence data represent the mixture of DNA molecules extracted, not automatically a pure population of the cells of interest.
Extraction methods influence DNA yield, fragment length, and representation. Short-read sequencing can work with relatively fragmented DNA, whereas many long-read workflows benefit from high-molecular-weight molecules. Mechanical shearing, repeated freezing and thawing, and harsh extraction conditions can shorten DNA fragments. Formalin-fixed tissue presents additional challenges because fixation can cause cross-linking, fragmentation, and chemical changes that generate sequencing artifacts. Ancient DNA is typically highly fragmented and chemically damaged, requiring methods that distinguish authentic sequence information from damage and modern contamination.
Library preparation converts extracted DNA into molecules compatible with the sequencing platform. In a typical short-read workflow, DNA is fragmented and adapters are attached to the fragment ends. Adapters provide instrument-binding and sequencing-primer sites and may include sample indexes, allowing multiple libraries to be sequenced together. Some workflows use polymerase chain reaction amplification to increase library quantity; others avoid it when enough input DNA is available. Amplification can introduce uneven representation and duplicate fragments, particularly in regions with extreme base composition. PCR-free preparation reduces some of these biases but does not remove all sources of uneven coverage.
Experimental design determines how much sequence is needed and which comparisons will be informative. Routine germline human WGS often targets an average depth around 30-fold, but this is not a universal sufficiency threshold. Tumor studies, low-frequency mosaic variants, microbial mixtures, and polyploid organisms can require different depths and analytical strategies. A parent–child trio can help distinguish inherited variants from newly arisen changes. Paired tumor and non-tumor samples help separate acquired tumor variants from inherited variation, although each sample still requires careful evaluation of its cellular composition.
Controls and sample tracking are essential because sequencing errors are only one source of incorrect conclusions. Mislabeling, cross-sample contamination, and unexpected relatedness can affect otherwise technically successful experiments. Laboratories therefore assess DNA quantity and integrity, monitor contamination, and compare genetic fingerprints when appropriate. Biological replication and sampling across tissues or time points address different questions from additional sequencing depth: reading the same DNA more often cannot establish whether a finding is present in another tissue or representative of an entire population.
Sequencing platforms and molecular readout
Most contemporary WGS uses either short-read massively parallel sequencing or long-read sequencing. These categories describe the length of individual observations, but platforms also differ in molecular preparation, error patterns, throughput, and ability to preserve information about native DNA modifications. Read length is especially important when a genome contains repeated sequences longer than the reads themselves. If identical repeats occur at several locations, a short observation from within a repeat may not reveal which copy it came from.
A widely used short-read approach is sequencing by synthesis. Adapter-bearing DNA fragments are attached to a surface and amplified into clusters, each containing many copies derived from an original fragment. Fluorescently labeled nucleotides are incorporated in cycles, and imaging identifies the incorporated base at each cluster. Paired-end sequencing reads both ends of a fragment; a common configuration produces two reads of 150 bases each. Their expected orientation and approximate separation provide information beyond either read alone, assisting alignment and the detection of some insertions, deletions, and rearrangements.
Single-molecule real-time sequencing observes DNA synthesis by individual DNA polymerase molecules. In circular consensus sequencing, the same insert is traversed repeatedly, allowing multiple observations to be combined into a highly accurate consensus read. High-fidelity reads commonly extend over many thousands of bases, making them useful for resolving duplicated genes, repeat expansions, and complex chromosome structures. Their value arises from the combination of length and accuracy, rather than length alone.
Nanopore sequencing uses a different physical signal. A DNA molecule passes through a nanoscale pore, and changes in electrical current depend on the sequence occupying the pore. Computational basecalling converts the signal into nucleotide predictions. Because the method reads native molecules without requiring synthesis of a complementary strand during measurement, read length can extend to hundreds of thousands of bases under suitable preparation conditions. Electrical signals can also contain information about modifications such as DNA methylation, although modification calling requires appropriate models and validation.
No platform is uniformly best for every task. Short reads offer established workflows and high accuracy for many small-variant applications, while long reads improve access to repetitive and structurally complex regions. Performance also depends on DNA quality, sequence context, software, and the variant classes being evaluated. Hybrid strategies combine complementary data types: accurate reads may support base-level consensus, very long reads may bridge repeats, and long-range mapping information may help organize sequences into chromosomes. An instrument’s advertised output does not by itself establish that a particular biological question has been answered.
Computational reconstruction and variant discovery
The computational workflow begins by translating instrument signals into reads with associated quality information. A common measure is the Phred quality score,
where is the estimated probability that a base call is incorrect. A score of 30 corresponds to an estimated error probability of , or one error per thousand calls. These scores are statistical estimates, not guarantees about individual bases, and their calibration depends on the sequencing and basecalling system. Initial processing can also separate indexed samples, identify adapter sequence, and detect poor-quality or contaminated data.
In reference-based analysis, reads are aligned to a reference genome. Alignment algorithms search for genomic locations compatible with each read while allowing mismatches and gaps. Repeated sequences can produce several nearly equivalent locations, making mapping confidence distinct from basecalling confidence. A perfectly read fragment may still be placed incorrectly. Reference choice also matters: sequences absent from the reference can fail to align, align incompletely, or be assigned to a similar region elsewhere. Graph-based references and pangenomes provide ways to represent alternatives, although their data structures and analysis conventions are more complex than those of a single linear sequence.
Variant callers evaluate the aligned evidence to infer the sample’s sequence differences. For a single-base change, they consider the observed bases, base qualities, alignment quality, read orientation, and other features. Local assembly can reconstruct candidate sequences around insertions and deletions that are difficult to describe through individual read alignments. Calling a diploid genotype requires distinguishing two chromosome copies, whereas polyploid samples or mixed cell populations require different assumptions. Many modern callers use machine learning, but their performance still depends on how closely training and evaluation data resemble the samples being analyzed.
In de novo assembly, algorithms reconstruct longer sequences from overlapping reads or related representations of shared sequence. Short-read assemblers often use graphs built from fixed-length sequence words; long-read assemblers commonly exploit overlaps or assembly graphs adapted to long observations. Repeats, sequencing errors, heterozygosity, and contamination can create ambiguous paths. Assembly continuity is often summarized by statistics such as N50, the sequence length at which half the assembly lies in sequences of that length or longer. A high N50 does not establish correctness: erroneous joins can increase apparent continuity while reducing biological accuracy.
Downstream processing turns sequence differences into interpretable records. Common formats include FASTQ for reads and qualities, BAM or CRAM for alignments, and VCF for variants. Annotation connects variants to genes, transcripts, predicted molecular consequences, and previously observed frequencies. Reproducibility requires recording software versions, parameters, reference assemblies, and annotation releases because each can alter the result. Coordinate conversion between reference versions is not always exact, especially around rearrangements and newly resolved sequences, so the reference assembly forms part of the meaning of a reported genomic position.
Coverage, variant classes, and analytical limitations
Sequencing depth describes how many observations support the genome being analyzed. If reads have average length and the genome length is , nominal mean coverage is
For paired-end data, both reads contribute to the total number of sequenced bases. A mean depth of 30-fold means that the total read output is approximately 30 times the genome length, not that every position has exactly 30 observations. Library duplicates, uneven sampling, and reads that cannot be confidently aligned reduce the amount of usable information.
Under an idealized model of independent, uniform sampling, the probability that a position receives no reads is approximately . Real sequencing violates these assumptions. Regions with unusual base composition, long repeats, highly similar duplicated genes, and difficult DNA structures can remain poorly covered even at substantial average depth. Coverage breadth—the proportion of the genome reaching a specified usable depth—is therefore distinct from mean coverage. A clinically relevant gene may require separate assessment even when whole-genome summary metrics appear satisfactory.
Different variant classes demand different evidence. Single-nucleotide variants alter one base, while small insertions and deletions add or remove short stretches. Structural variants include larger deletions, insertions, inversions, translocations, and complex rearrangements. Copy-number variation changes the number of copies of a genomic segment. Read-depth shifts can suggest copy-number changes, discordant read pairs can indicate altered spacing or orientation, and split reads can span breakpoints. Long reads may traverse entire rearrangements or repeat expansions that short reads cannot resolve directly.
Assigning variants to chromosome copies is another challenge. A haplotype is a set of variants carried together on one chromosome. Two variants in the same gene can have different implications depending on whether they occur on the same copy or on opposite copies. Statistical phasing uses population patterns, while read-based phasing uses molecules spanning multiple variants; family data can also establish inheritance. Complete chromosome-scale phasing generally requires more information than a list of unphased small variants.
Low-frequency variation creates further limits. Mosaicism means that genetically different cell populations occur within an individual, so a variant may appear in only a minority of sampled DNA molecules. A negative result in blood does not exclude a variant confined to another tissue. Similarly, mitochondrial variants can occur at different proportions among mitochondrial DNA molecules, a condition called heteroplasmy. Analytical sensitivity depends on depth, error rates, sample composition, and filtering thresholds. Validated performance is therefore expressed for specified variant classes and conditions, rather than as a single claim that a genome has been completely examined.
Human disease research and clinical use
WGS can identify genetic changes associated with disease, but detecting a variant is not equivalent to establishing a diagnosis. Human genomes contain millions of differences from a reference, most of which do not cause a highly penetrant disorder. Clinical interpretation combines molecular evidence with inheritance patterns, population frequency, laboratory findings, and the patient’s phenotype. A rare variant may be incidental, and a variant predicted to disrupt a gene may have different consequences depending on the transcript, tissue, and underlying disease mechanism.
In rare-disease investigation, WGS can detect coding changes as well as some variants missed by exome sequencing, including deep intronic changes, structural rearrangements, and variants in poorly captured regions. Sequencing affected individuals and their parents can reveal newly arisen variants, recessive inheritance, or segregation patterns that narrow the search. Rapid sequencing has also been used in critically ill infants where a molecular result can clarify an otherwise uncertain disorder. Diagnostic yield varies substantially with patient selection, previous testing, available family samples, and the scope of analysis; it is not a fixed property of the technology.
Interpretation systems distinguish degrees of evidence rather than forcing every finding into “causal” or “harmless.” The American College of Medical Genetics and Genomics and the Association for Molecular Pathology published widely used sequence-variant interpretation guidelines in 2015, including the categories pathogenic, likely pathogenic, uncertain significance, likely benign, and benign. These categories concern evidence for a variant’s relationship to disease in an appropriate context. A variant of uncertain significance is not an established explanation of symptoms. Reanalysis can change an interpretation when gene–disease relationships, population data, or functional evidence become available.
In cancer, WGS investigates acquired mutations, chromosome rearrangements, copy-number alterations, and patterns of DNA damage and repair. Comparing a tumor with a matched non-tumor sample helps distinguish somatic changes from inherited variants. Tumor purity, subclonal diversity, and abnormal chromosome number complicate inference: a variant’s observed fraction among reads does not directly equal the fraction of tumor cells carrying it. Genome-wide rearrangement patterns and mutational signatures can illuminate tumor biology, but not every detected alteration is clinically actionable or informative about treatment response.
Other applications include pharmacogenomics, in which inherited variation affects drug metabolism or response, and research on common diseases using genome-wide association studies. Sequencing can reveal rare alleles and structural variants beyond those captured by genotyping arrays. Nevertheless, technical detection, disease association, risk prediction, and demonstrated clinical utility are separate standards. Genetic counseling addresses the familial implications and uncertainty of genomic findings, while laboratory reports distinguish established findings from research observations and acknowledge relevant analytical limitations.
Microbial surveillance, evolution, and biodiversity
Microbial WGS can provide high-resolution comparisons among organisms that are difficult to distinguish using a small number of laboratory markers. For many bacteria, genomes contain only a few million base pairs, making large collections comparatively tractable to sequence. Analyses can identify species, distinguish lineages, characterize accessory genes, and detect plasmids or other mobile elements. Unlike a single marker-gene assay, WGS can evaluate both shared core sequence and variable genomic content, although recovery of every mobile element depends on sample preparation and assembly quality.
In public health, genome comparisons help investigate outbreaks of foodborne and healthcare-associated infections. Closely related isolates may support a hypothesis of recent shared ancestry, while substantial genomic differences can argue against a single transmission cluster. However, genomic similarity alone does not prove that one person infected another or identify the precise route of transmission. Interpretation requires sampling dates, exposure information, knowledge of background diversity, and the organism’s evolutionary behavior. No universal number of single-nucleotide differences defines an outbreak across all microbial species.
WGS also identifies genetic determinants associated with antimicrobial resistance, such as resistance genes or mutations in drug targets. It can reveal whether resistance is associated with a spreading lineage or with mobile DNA moving between lineages. Horizontal gene transfer is particularly important because a resistance plasmid can disseminate across otherwise distinct bacterial backgrounds. Genotypic prediction does not always replace laboratory susceptibility testing: gene expression, novel mechanisms, regulatory changes, and incomplete knowledge can produce disagreement between an organism’s sequence and its measured response to an antibiotic.
In evolutionary research, whole genomes provide large numbers of characters for phylogenetic reconstruction and the study of population genetics. They can reveal population structure, migration, recombination, and changes in effective population size. Comparisons across species identify conserved sequences and lineage-specific changes. In humans and other organisms, distinguishing natural selection from demographic history and genetic drift requires explicit models because different processes can generate similar genomic patterns. A large dataset does not remove ambiguity if the evolutionary assumptions are inappropriate.
WGS contributes to biodiversity research by generating reference genomes, documenting genetic diversity, and examining adaptation in threatened populations. Long-read assemblies can expose chromosome rearrangements and duplicated regions previously hidden in fragmented references. Sequencing environmental mixtures is usually described as metagenomics rather than WGS of a single organism, because DNA from many organisms is analyzed together. Likewise, sequencing RNA viruses generally requires an RNA-to-DNA or other RNA-compatible workflow; it is genome sequencing, but not the same laboratory procedure as conventional DNA-based WGS. These distinctions determine which organisms and genomic components an experiment can detect.
Data governance, representation, and continuing challenges
Human genome data are both individually identifying and relational. A sequence contains information about biological relatives as well as the person sampled, and portions of it can be informative about ancestry or disease susceptibility. Removing names and direct identifiers reduces some risks but does not guarantee anonymity, particularly when genomic data can be linked with genealogical or other datasets. Data privacy therefore involves access controls, security, permitted uses, and institutional accountability rather than reliance on de-identification alone.
Informed consent for sequencing addresses the intended analysis, storage, sharing, and possible future use of samples and data. Research and clinical sequencing differ in their purposes and obligations, especially concerning whether findings are returned to participants. Secondary findings are medically relevant results outside the original reason for testing; incidental findings arise unexpectedly during analysis. Policies vary across institutions and jurisdictions, and professional recommendations evolve. Consent processes also confront practical questions about recontact, reinterpretation, research use after a participant’s death, and findings relevant to relatives.
Representation affects scientific validity. Historically, many genomic reference resources and association studies have included disproportionate numbers of people with European ancestry. When comparison data are sparse for a population, an uncommon benign variant may be harder to distinguish from a disease-associated one. Risk prediction models can also perform differently across populations because allele frequencies, linkage patterns, environmental conditions, and study designs differ. Broader sampling improves the evidence base, but categories of ancestry remain imperfect descriptions of continuous and complex population histories.
Technical progress creates additional governance and infrastructure demands. Long reads and improved assemblies increase the amount of interpretable sequence, while pangenomes challenge systems built around a single reference coordinate framework. Reanalysis may require access to original reads, not merely a final variant list, because new algorithms can recover signals that earlier pipelines missed. Long-term storage decisions consequently affect future scientific utility, reproducibility, and exposure to unauthorized access. The costs of sequencing instruments and reagents are only part of the total; sample handling, computation, validation, interpretation, and secure data stewardship remain substantial components.
A central unresolved problem is the interpretation of noncoding and structurally complex variation. Sequence conservation, molecular assays, population observations, and computational predictions provide complementary evidence, but none alone establishes the consequence of every variant. Functional studies may test a regulatory sequence outside its native chromosomal context, while statistical association can implicate a region without identifying the causal change. Accurate genome reconstruction supplies the substrate for these investigations; connecting sequence to cellular behavior, development, and disease requires additional experimental and clinical evidence.