Exome sequencing is a form of DNA sequencing that preferentially analyzes the protein-coding regions of a genome, together with varying amounts of adjacent sequence. These regions collectively constitute the protein-coding exome. In humans, coding sequence accounts for approximately 1–2% of nuclear DNA, yet contains many variants with established roles in inherited disease. By concentrating sequencing effort on this comparatively small fraction of the genome, exome sequencing makes broad investigation of coding variation possible without sequencing every genomic region to equivalent depth.
The method became established in the late 2000s and has since become an important tool for identifying disease-associated genes, investigating rare disorders, characterizing cancers, and studying variation in large populations. Its central strength is breadth across coding genes rather than exhaustive coverage of every type of genetic alteration. Generating sequence data is only one stage: laboratory enrichment, computational analysis, assessment of inheritance, and interpretation of biological evidence all influence what the test can establish.
Biological scope and historical development
An exon is a segment of a gene retained in a mature RNA molecule after RNA splicing. Exons are not necessarily entirely protein-coding: they can also contain untranslated regions, and some belong to noncoding transcripts. Consequently, “whole-exome sequencing” is an operational term rather than a guarantee that every exon in every transcript has been analyzed. Most human exome assays primarily target annotated coding exons, often adding nearby splice junctions and selected untranslated or clinically important noncoding regions. Their precise contents depend on the capture design and the transcript annotations used.
Human coding exons together occupy roughly 30 million base pairs, although commercial target intervals can be substantially larger because they include flanking sequences and additional annotations. The distinction between target size and coding size matters when comparing platforms. A design covering 50 million bases does not necessarily cover more protein-coding sequence than a smaller design; some of its additional sequence may lie outside coding regions. Furthermore, alternative splicing means that the same genomic position may be coding in one transcript and intronic in another.
The biological rationale for exome sequencing rests partly on how coding changes affect proteins. A nucleotide substitution may replace an amino acid, introduce a premature termination signal, or leave the encoded amino acid unchanged. Insertions and deletions may disrupt the reading frame. Changes at splice junctions can alter exon inclusion or intron removal. These effects often provide more direct interpretive clues than changes in poorly characterized noncoding regions, although neither a coding location nor a predicted protein alteration establishes that a variant causes disease.
Exome sequencing emerged from improvements in massively parallel sequencing and methods for enriching selected DNA fragments. Before these developments, identifying an unknown disease gene commonly required linkage mapping in large families followed by sequencing of positional candidates. Exome analysis made it feasible to inspect coding regions across thousands of genes simultaneously, including in small families and isolated cases.
A landmark 2009 study by Sarah Ng and colleagues demonstrated the approach by sequencing exomes from individuals with Freeman–Sheldon syndrome and recovering the known disease-associated gene, MYH3, through variant filtering. Work published in 2010 used exome sequencing to identify DHODH as a gene responsible for Miller syndrome. These studies helped establish a general strategy: sequence affected individuals, exclude common variants, consider inheritance, and investigate the remaining candidates. Later applications expanded from discovering highly penetrant causes of rare disorders to clinical diagnostics and population-scale analyses of coding variation.
Laboratory enrichment and sequencing
The laboratory workflow generally begins with DNA extracted from blood, saliva, cultured cells, or another tissue. Sample choice influences the biological question. Blood commonly represents constitutional variation, whereas tumor tissue contains mixtures of malignant and nonmalignant cells. Formalin-fixed material can be fragmented and chemically altered, creating additional challenges. Sample identity, DNA quantity, fragment integrity, and contamination are assessed before or during library preparation because deficiencies at this stage can propagate through the entire analysis.
A sequencing library is a collection of DNA fragments carrying adapters that permit amplification, identification, and sequencing. Genomic DNA is fragmented, its ends are prepared, and adapters are attached. Sample-specific index sequences allow several libraries to be pooled in one sequencing run. Some protocols incorporate unique molecular identifiers, which label original molecules and can help distinguish independent observations from copies produced during amplification. Library preparation may use polymerase chain reaction, although amplification conditions and the number of cycles vary among protocols.
Most conventional exome workflows use hybridization capture. Short synthetic DNA or RNA probes, also called baits, are designed to complement target regions. Library fragments hybridize to these probes, and the resulting complexes are isolated, commonly using biotinylated probes and streptavidin-coated magnetic beads. Washing removes many unbound fragments, after which the enriched material is amplified if necessary and sequenced. Capture enriches fragments rather than cutting out exact exon boundaries, so reads often extend into neighboring intronic sequence.
Alternative enrichment methods use multiplex amplification to generate products from predefined regions. These approaches can work with limited DNA and support specialized applications, but large numbers of primer pairs introduce design constraints and amplification biases. In hybridization capture, enrichment efficiency likewise varies with sequence composition, fragment length, probe design, and hybridization conditions. Regions with unusually high or low guanine–cytosine content and regions closely resembling other genomic sequences can be particularly difficult.
Sequencing is usually performed with short-read instruments, often producing paired-end reads from both ends of each fragment. Clinical germline exomes frequently aim for mean target depths around 80–150-fold, although laboratory specifications differ and average depth alone does not define test quality. Tumor analyses or investigations of low-frequency variants may require greater depth. A nominally deep experiment can still leave particular exons inadequately covered.
Laboratory quality assessment therefore includes the fraction of reads aligning to the genome, the fraction falling on target, duplicate levels, coverage uniformity, and the proportion of target bases exceeding specified depth thresholds. Capture batches and reagent versions can change these properties. Meaningful comparison between samples requires attention to the actual regions reliably interrogated, not merely the total number of reads produced.
Computational processing and variant detection
Computational analysis converts instrument output into a set of candidate differences from a reference sequence. Initial processing assesses base-call quality, adapter contamination, and sample indexing. Reads are then aligned to a reference genome. Widely used human references include GRCh37 and GRCh38, but coordinates, gene annotations, and variant descriptions must remain consistent throughout the analysis. A variant reported without a reference assembly and transcript context may be ambiguous, especially where transcript structures or reference sequences differ.
Alignment places each read at its most plausible genomic origin and assigns a mapping-quality estimate. This is straightforward for many unique regions but difficult where sequences are duplicated. A read derived from a pseudogene may align to a functional gene, or the reverse. Read-pair relationships and local sequence context help resolve some ambiguities, but not all. Duplicate marking identifies fragments likely to represent repeated sequencing of the same original molecule, preventing amplification from being mistaken for independent supporting evidence.
Variant calling then evaluates the aligned evidence for substitutions and small insertions or deletions. Contemporary callers often reconstruct local sequence arrangements rather than examining each position independently, improving the handling of nearby changes and short indels. Germline callers estimate genotypes such as homozygous reference, heterozygous, or homozygous alternate, typically using probabilistic models of base quality, read support, and alignment uncertainty. Joint analysis across samples can improve consistency and identify sites with recurrent technical problems.
Mean sequencing depth over a specified target can be expressed as
where is the number of aligned sequenced bases overlapping that target and is its length. This average conceals unevenness. Variant detection also depends on the number of independent molecules, strand balance, mapping quality, and the fraction of reads supporting the alternate sequence. A heterozygous constitutional variant often has an alternate-read fraction near one half, but sampling variation and technical biases can shift that fraction.
Exome data can also reveal some copy-number variants through relative read depth across exons. Such analysis normally compares a sample with suitably matched controls because capture efficiency differs substantially among targets. Multi-exon deletions and duplications may be detectable, whereas single-exon events and complex rearrangements are more challenging. Breakpoints usually lie outside the captured regions, limiting structural resolution.
Finally, variants are annotated with gene names, transcript consequences, population frequencies, disease associations, and computational predictions. Normalization ensures consistent representation of changes that can be described in more than one equivalent way. Quality filters remove or flag unreliable calls, but no universal threshold suits every context. Low-frequency tumor variants, for example, require different assumptions from inherited heterozygous variants. Reproducible analysis records software versions, reference resources, filtering rules, and the regions excluded from confident interpretation.
Interpreting variants and establishing causality
A typical human exome contains tens of thousands of differences from the reference sequence. The interpretive task is not simply to find a difference, but to determine whether particular variants explain a biological observation or clinical phenotype. Common filters consider population frequency, predicted molecular consequence, known gene–disease relationships, inheritance, and the fit between a person's features and previously established disease manifestations. Each filter expresses an assumption, and overly restrictive assumptions can discard the true explanation.
Inheritance provides especially useful structure. In an autosomal recessive disorder, candidate explanations may involve a homozygous variant or two relevant variants on opposite chromosome copies. Establishing whether two variants are in trans, rather than on the same haplotype, can be decisive. Dominant disorders may involve an inherited variant with variable expression or a de novo change absent from parental samples. X-linked analysis requires attention to chromosomal context, sex chromosome complement, and the relevant disease mechanism. Parental testing can clarify these relationships and help detect sample or pedigree inconsistencies.
Trio sequencing analyzes an affected individual and both biological parents together. It supports identification of de novo variants, recessive inheritance, and parental origin. However, an apparently de novo call still requires technical scrutiny: inadequate coverage in a parent or low-level parental mosaicism can alter the interpretation. Conversely, finding a variant in an unaffected parent does not automatically exclude causality because some disorders show incomplete penetrance, age-dependent expression, or variable severity.
Clinical laboratories commonly organize evidence using the five-category framework published in 2015 by the American College of Medical Genetics and Genomics and the Association for Molecular Pathology: pathogenic, likely pathogenic, uncertain significance, likely benign, and benign. Classification combines multiple evidence types, including population frequency, segregation, functional studies, and established molecular mechanisms. Disease-specific expert specifications can refine how individual criteria are applied.
A variant of uncertain significance is not an established molecular diagnosis. Rarity alone is weak evidence because every person carries rare coding variants. Likewise, computational predictions are supporting information rather than direct demonstrations of functional harm. A protein-truncating change is not necessarily pathogenic: its significance depends on whether loss of function causes the disease, whether the affected transcript is relevant, and whether the position permits escape from RNA surveillance or preservation of essential function.
Interpretation also distinguishes variant pathogenicity from relevance to a particular person. A clearly pathogenic variant may be unrelated to the presenting features, or a single pathogenic allele may indicate carrier status for a recessive condition. Functional experiments, segregation in additional relatives, and independent cases can strengthen a proposed relationship. When the gene itself lacks an established disease association, even compelling sequence findings require careful separation of research hypotheses from clinically validated conclusions.
Clinical applications and diagnostic workflows
Exome sequencing is used clinically to investigate disorders for which many different genes can produce similar features. Examples include developmental disorders, intellectual disability, epilepsy, congenital anomalies, inherited muscle disease, and heterogeneous metabolic conditions. It can also help evaluate atypical presentations in which selecting a narrow list of candidate genes is difficult. Its value arises from testing many plausible coding explanations simultaneously rather than requiring a sequence of individually chosen gene tests.
Diagnostic yield is the proportion of tested individuals for whom an explanatory result is established under specified criteria. Published clinical exome cohorts commonly report yields of approximately 25–40%, but this is not a universal performance characteristic. Yield varies with referral population, previous testing, family structure, phenotype quality, analytical methods, and the definition of a positive finding. Highly selected families with a strong indication of Mendelian inheritance may differ markedly from broadly referred clinical populations.
A 2021 evidence-based guideline from the American College of Medical Genetics and Genomics supported considering exome or genome sequencing as a first- or second-tier test for pediatric patients with congenital anomalies, developmental delay, or intellectual disability. This recommendation concerned defined clinical groups rather than unrestricted population screening. Clinical implementation still depends on laboratory validation, appropriate interpretation, and the capabilities of the surrounding health service.
Phenotypic information is central to analysis. Structured descriptions can use the Human Phenotype Ontology, which represents findings such as seizures, short stature, or particular skeletal abnormalities in a standardized vocabulary. Detailed positive and negative findings, age at onset, and family history help prioritize genes. Automated matching can rank candidates, but resemblance scores do not establish causation, and newly recognized disorders may not yet have comprehensive phenotype annotations.
A molecular diagnosis can clarify disease classification, distinguish conditions with overlapping clinical appearances, and identify established implications for surveillance, treatment, or reproductive risk. These consequences depend on the specific disorder and the strength of the result. Some individuals receive more than one molecular diagnosis, accounting for a blended presentation that no single condition fully explains. Others have a relevant diagnosis despite features that differ from the initially recognized spectrum.
An uninformative result does not exclude a genetic disorder. Clinical workflows may incorporate review of poorly covered genes, analysis of variant classes not reliably captured by exome sequencing, or later reanalysis. Reanalysis can identify explanations when new gene–disease associations become established, transcript annotations improve, or a person's phenotype evolves. Its yield depends on the original methods and elapsed time; it is not simply a repetition of the initial laboratory sequencing. Reports therefore distinguish the findings obtained from the boundaries of what the assay and interpretation could assess.
Gene discovery, population studies, and cancer research
In rare-disease research, exome sequencing supports discovery by finding convergent evidence across unrelated families. A single family may produce a plausible candidate gene, but independent individuals with comparable phenotypes and variants affecting the same biological mechanism provide stronger evidence. International case matching has been particularly important for disorders too rare for one center to assemble a substantial series. Experimental studies can then investigate whether the implicated changes alter protein activity, cellular localization, development, or other relevant processes.
Exome analysis is also used to identify de novo enrichment across groups of affected individuals. Rather than treating every newly arising variant as equally informative, statistical models account for gene length, sequence context, and background mutation rates. Repeated disruptive changes in the same gene may support association when their observed number exceeds expectation. Such studies require attention to multiple testing and to whether cases and comparison data were processed in compatible ways.
Large population projects have transformed interpretation by measuring how frequently coding variants occur in different ancestry groups. Resources such as the Genome Aggregation Database aggregate extensive exome and genome data to estimate allele frequencies and patterns of coding constraint. If a variant is substantially more common than a severe, highly penetrant rare disorder would permit, that frequency argues against a simple causal interpretation. Conversely, absence from a database does not establish pathogenicity, particularly for underrepresented populations.
Population exomes also support rare-variant association studies. Because most rare variants occur in too few individuals for reliable individual testing, researchers often aggregate variants within a gene. Burden tests ask whether qualifying variants collectively occur more often among cases than controls; other methods allow effects to differ in direction or magnitude. Choices about which variants qualify—such as predicted loss-of-function changes or a subset of missense substitutions—can strongly influence results. These analyses complement genome-wide association studies, which have traditionally emphasized common variation.
In cancer research, exome sequencing frequently compares tumor DNA with matched constitutional DNA. This helps separate acquired somatic changes from inherited variants. Analysis must account for tumor purity, mixtures of subclones, copy-number alterations, and contamination. The fraction of reads supporting a somatic change is therefore not a simple indicator of its cellular prevalence.
Tumor exomes have helped identify recurrently altered genes, mutational processes, and candidate neoantigens. However, most observed somatic changes are passengers rather than drivers of tumor growth. Estimates of tumor mutational burden also depend on the analyzed territory, variant filters, sequencing quality, and inclusion or exclusion of particular mutation classes. Results obtained from a research exome and a smaller clinical panel are not automatically interchangeable. Moreover, some clinically important rearrangements and noncoding alterations lie outside the regions that exome sequencing interrogates well.
Analytical limitations and comparison with other methods
Exome sequencing is selective by design. Deep intronic variants, distant regulatory changes, many repeat expansions, and balanced chromosomal rearrangements generally fall outside its strongest capabilities. Certain alterations may leave indirect clues in exome reads, but incidental detectability differs from a validated analytical claim. A negative exome result therefore means that no explanatory finding was established within the assay's effective scope, not that all possible genetic mechanisms were excluded.
Even targeted coding regions are not uniformly accessible. Repeated sequences, high sequence similarity between genes and pseudogenes, extreme base composition, and alignment ambiguity can reduce sensitivity. A capture probe may enrich both a gene and its close relatives, while short reads may not reveal which locus produced the sequence. Genes such as PMS2, with closely related pseudogene sequence in clinically relevant regions, illustrate why specialized methods may be necessary for particular loci.
Sensitivity also depends on biological sampling. A variant present only in a small proportion of cells may produce too few supporting reads to distinguish it from error. Tissue-restricted mosaicism can be absent from blood even when it is abundant in an affected organ. Mitochondrial variants introduce additional issues because the proportion of altered mitochondrial genomes, called heteroplasmy, differs among tissues and changes over time. Off-target mitochondrial reads from a standard exome do not necessarily provide validated mitochondrial genome analysis.
Compared with a targeted gene panel, an exome offers broader scope and the possibility of revisiting additional genes without new sequencing. Panels can nevertheless achieve deeper, more uniform coverage and incorporate specialized analysis for difficult genes or variant types. Some laboratories generate broad exome data but initially interpret only a predefined gene set, sometimes called a virtual panel. This separates the scope of data acquisition from the scope of interpretation.
Whole-genome sequencing avoids the exon-capture step and surveys both coding and noncoding regions. It often provides more uniform coding coverage and better access to structural variation and intronic changes. However, standard short-read genome sequencing still has difficulty with some repeats, complex rearrangements, and highly homologous loci. Its larger search space also increases computational demands and leaves many noncoding findings difficult to interpret. Relative costs depend on sequencing scale, storage, analysis, and local infrastructure rather than reagent prices alone.
Other methods address different questions. RNA sequencing can reveal abnormal splicing or transcript abundance, but relevant genes may not be expressed in the sampled tissue. Long-read sequencing can improve phasing and resolve some repetitive or structurally complex regions. Chromosomal microarrays, repeat-specific assays, and methylation analysis remain informative for variant classes that exomes do not comprehensively assess. The methods are therefore partly complementary rather than points on a single scale from less to more complete.
Consent, data stewardship, and equitable implementation
Exome sequencing can produce information beyond the immediate reason for testing. A primary finding addresses the presenting question, while a secondary finding results from deliberate analysis of specified additional genes. An incidental finding is encountered unexpectedly during the investigation. These categories overlap in ordinary usage, but their distinction matters for consent, laboratory policy, and the scope of reporting. A broad dataset does not imply that every potentially relevant gene has been examined or that every technically detectable result will be disclosed.
Professional organizations have developed policies for reporting selected secondary findings associated with serious conditions for which established medical interventions exist. The American College of Medical Genetics and Genomics first issued such recommendations in 2013 and has revised its gene list and implementation policies over time. Policies vary across countries and institutions, including how choices about secondary analysis are recorded and how testing of children is handled. Reporting secondary findings remains distinct from claiming that exome sequencing constitutes comprehensive population screening.
Consent processes address the purposes of sequencing, possible result categories, storage and reuse of samples, data sharing, and potential implications for relatives. Family-based analysis can uncover unexpected biological relationships or inheritance patterns. Exome information is also not anonymous in an absolute sense: combinations of genetic variants can be identifying, especially when linked with other datasets. Data privacy therefore involves access controls, secure transfer, governance of research reuse, and limits on the distribution of identifiable clinical information.
Data stewardship extends beyond keeping a sequence file. Clinical interpretation depends on links among sample identity, phenotype, reference assembly, transcript versions, analytical software, and the evidence available at a particular time. A result may need revision as gene–disease knowledge develops. Institutions differ in their arrangements for reanalysis, amended reports, and recontact. Maintaining these connections requires continuing organizational resources; storing raw data alone does not preserve the full interpretive context.
Equity affects both access and accuracy. Underrepresentation of some ancestry groups in reference datasets can leave more variants classified as uncertain because there is less evidence about their background frequencies. Unequal access to specialist assessment, family testing, and follow-up can further influence who receives an interpretable diagnosis. These differences cannot be resolved solely by increasing sequencing throughput. More representative datasets, appropriately governed collaboration, and broader clinical expertise all affect the usefulness of an exome result.
Implementation costs likewise include more than generating reads. Clinical assessment, interpretation, confirmation where required, reporting, information systems, and long-term data management can represent substantial parts of the service. Comparisons with other testing strategies depend on whether they count these activities and downstream consequences. As sequencing technologies change, the continuing operational question is which combinations of assay scope, analytical validation, and interpretive capacity can reliably address a defined clinical or research problem.