Genome Interpretation
How Genomic Data Becomes a Clinical Diagnosis

Genome interpretation, also called variant interpretation, is the process of evaluating genetic variants identified during sequencing to determine their clinical significance. It begins where computational analysis ends, at a list of variants in a VCF file, each annotated with biological context, and ends at a defensible clinical conclusion about which variants cause or contribute to the patient's condition.

The four questions for every candidate variant

• Is the variant real? Quality assessment: is this a true variant or a sequencing artifact?
• Is it rare enough to be disease-causing? Population frequency in the relevant ancestral group.
• Does it affect gene function? Effect on protein, transcript, or expression.
• Does it fit the clinical picture? Gene-disease relationship matches the phenotype, inheritance fits the family history.

Every variant that reaches clinical reporting has been evaluated against all four. The answers, weighted by strength of evidence, determine the final classification. Genome interpretation sits within the broader NGS pipeline as the clinical component of tertiary analysis. For the upstream steps (annotation, filtering, prioritization) that precede formal classification, see the tertiary analysis guide.

The clinical context shapes interpretation. The most important contextual factor is what type of genetic condition is being investigated. The four types differ in inheritance, genetic architecture, and penetrance, and therefore in how variants are interpreted and weighted.

Genome interpretation is fundamentally an evidence-weighing exercise. No single piece is sufficient. Pathogenicity conclusions are built by integrating multiple independent lines of evidence, each contributing different information about biological and clinical significance. The ACMG/AMP framework organizes evidence into the following categories, each with a defined weight.

The 2015 ACMG/AMP variant classification framework (Richards et al., Genetics in Medicine) is the global standard for germline variant interpretation. It defines 28 criteria organized by evidence type and strength, and rules for combining them to reach a 5-tier classification.

The five tiers

• Pathogenic (P): the variant causes disease. Clinical management should be guided by this finding. Typically requires either PVS1 plus supporting evidence, or multiple Strong criteria.
• Likely Pathogenic (LP): probability of disease-causing exceeds 90%. Clinically managed the same as Pathogenic in most contexts, with appropriate disclosure of remaining uncertainty.
• Variant of Uncertain Significance (VUS): insufficient evidence to classify. Cannot guide clinical management. See the dedicated VUS section below.
• Likely Benign (LB): probability of being benign exceeds 90%. Not reported in most clinical contexts but retained in the lab database for future reclassification monitoring.
• Benign (B): the variant does not cause disease. Established by very high population frequency (BA1) or multiple independent lines of benign evidence.

Key criteria explained

Combining criteria

The framework provides explicit combination rules. Pathogenic classification requires combinations such as: 1 Very Strong + 1 Strong; 1 Very Strong + 2 Moderate; 2 Strong; 1 Strong + 3 Supporting; 2 Moderate + 2 Supporting (among others). Likely Pathogenic requires lower combinations. These rules ensure that classifications are reproducible across labs when criteria are applied consistently.

ClinGen Expert Panel refinements

For many clinically important genes, ClinGen Expert Panels have published gene-specific specifications that refine how ACMG criteria are applied: gene-specific allele frequency thresholds for PM2 and BS1, which functional assays meet PS3 strength, phenotype specificity thresholds for PP4, PVS1 application adjustments for genes with complex isoform structures. Gene-specific rule sets exist for BRCA1/2, MLH1/MSH2/MSH6/PMS2, SCN5A and KCNQ1, MYH7 and MYBPC3, and dozens of others. Using gene-specific specifications substantially reduces inter-laboratory variability and VUS rates.

Whether you are a clinician receiving a genomic report for a patient or an individual reviewing your own results, understanding the structure of a clinical genomics report is essential for translating the laboratory's findings into appropriate action.

Standard report structure

• Specimen and test information: patient ID, sample type, test name, assay methodology, sequencing platform, reference genome build, coverage metrics.
• Summary of findings: brief clinical summary of the most significant result.
• Variant section: each reported variant with HGVS nomenclature (genomic, coding, protein), gene, classification, zygosity, and supporting evidence summary.
• Clinical interpretation: how the identified variant(s) explain or relate to the presentation. For P or LP findings, connects the molecular result to phenotype and provides management context.
• Limitations: what the test did not cover, variant types not detected, known coverage gaps, caveats about negative results.
• Recommendations: confirmatory testing, specialist referral, cascade family testing, surveillance, therapeutic implications.

What each tier means for management

• Pathogenic: clinically significant. Management (surveillance, risk reduction, therapy selection, family testing) should be guided by this result in conjunction with the patient's full clinical picture and specialist consultation.
• Likely Pathogenic: treated clinically the same as Pathogenic in most contexts. Probability exceeds 90%. Acknowledge residual uncertainty when disclosing to patients.
• VUS: cannot guide clinical management. Management should be based on the clinical presentation, not the VUS finding. See dedicated VUS section.
• Likely Benign / Benign: not clinically actionable. The variant is very likely or definitively not causing disease in the gene where it was found.

Variants of uncertain significance are the most common result type in clinical genomic sequencing for many patient populations, and the most misunderstood. VUS rates of 30 to 50% in WES for unselected rare disease populations are typical: not because the technology is failing, but because genomic knowledge is genuinely incomplete for a large fraction of rare variants.

Why VUS rates are high

The human genome contains ~4 to 5 million variants relative to the reference in any individual. The vast majority have never been observed in a clinical context. Even for variants in disease-associated genes, most specific nucleotide changes have never been seen before in an affected patient. They are novel variants with no prior clinical report. Without prior clinical evidence, pathogenicity determination depends entirely on functional evidence, computational predictions, and population frequency. None alone is sufficient for a definitive classification. The combination of weak positive signals and no prior clinical reports produces a VUS.

What a VUS is not

A VUS is not a positive result. It should not trigger the same clinical response as a Pathogenic finding. Many clinicians and patients misinterpret a VUS as evidence of disease, leading to unnecessary surveillance, prophylactic procedures, and psychological distress in patients who may not carry a disease-causing variant at all.

A VUS is also not a negative result. The variant may eventually be reclassified as Pathogenic as evidence accumulates. The appropriate response is watchful waiting with systematic reclassification monitoring.

What labs and patients should do

• Labs: implement proactive reclassification monitoring. A systematic process for identifying VUS findings in the lab's patient cohort that have been reclassified in external databases and proactively notifying ordering clinicians. This is a patient safety function, not just an operational nicety.
• Clinicians: do not act on VUS findings clinically. Counsel patients about what a VUS means, document it, and establish a re-contact plan for when reclassification occurs.
• Patients: understand that a VUS does not provide an answer and is not the same as a positive result. Stay connected with the ordering lab so reclassification notifications can reach you.

Published studies consistently document VUS reclassification rates of 10 to 30% over 3 to 5 years, with the majority of reclassifications moving toward Benign or Likely Benign rather than Pathogenic. Important context for patients: most VUS findings eventually resolve to benign classifications.

Genome interpretation does not occur in a clinical vacuum. The same variant can have different clinical significance depending on the patient's phenotype, inheritance context, and family history. This is why variant interpretation is not purely a laboratory function. It requires genuine integration of molecular and clinical information.

Why phenotype drives interpretation

• Prior probability. A patient with confirmed Marfan features presenting for molecular confirmation has a very high prior probability that a variant in FBN1 is causative. The same variant in a patient with no connective tissue features has a much lower prior probability.
• Gene relevance. A candidate variant in a gene with no established association to the presentation cannot support a diagnosis regardless of functional impact. Conversely, a moderate-evidence variant in a gene with perfect phenotypic match may be clinically significant even without strong functional data.
• Inheritance pattern confirmation. A heterozygous variant in an autosomal recessive gene is not a diagnosis. It makes the patient a carrier. If the presentation is consistent with a recessive condition and no second variant is found, the interpretation must address whether a second hit has been missed (poor coverage, intronic variant, deletion) rather than incorrectly attributing the condition to the single heterozygous finding.

HPO terms and computational phenotype matching

Human Phenotype Ontology (HPO) terms provide a standardized vocabulary for clinical features that enables computational phenotype-genotype matching. When a patient's HPO terms are entered into an interpretation platform, gene-disease relationships can be ranked by phenotypic similarity, surfacing candidates in genes with established associations to the patient's specific feature combination. This is why thorough phenotypic documentation is a prerequisite for efficient interpretation. A report listing "intellectual disability" provides less information than one listing the specific features (dysmorphic facial features, absent speech, seizure onset at 6 months, hypotonia, brain MRI findings) that together create a distinctive phenotypic fingerprint pointing toward specific diagnoses.

Genome interpretation is improving rapidly, but several structural challenges affect clinical quality and consistency across labs. Each is being actively addressed, but each remains a real source of variability in clinical results today.

The volume of genomic testing has grown faster than the workforce of clinical geneticists and molecular pathologists capable of performing manual interpretation. Automation is not optional. It is necessary for clinical genomics to scale. The question is not whether to automate but where automation is appropriate and where human judgment remains essential.

Where automation adds value

Evidence gathering is the most automatable component. Pulling allele frequencies from gnomAD, retrieving ClinVar classifications, running in silico prediction tools, calculating SpliceAI scores, and searching literature databases: all can be automated reliably and produce the same results as manual database queries, faster and without transcription errors.

ACMG criteria application is partially automatable. PVS1 (null variant assessment), BA1 (high frequency), PM2 (absent from population databases), and PP3/BP4 (computational predictions) can all be applied algorithmically with high accuracy when properly calibrated. Many criteria, including PS3 (functional evidence quality), PP4 (phenotype specificity), and PM3 (trans configuration in recessive disease), require human judgment that cannot be reliably automated.

Clinical interpretation synthesis (translating the molecular finding into a clinical report that accurately characterizes the variant's significance for the specific patient) remains a human responsibility. No algorithm can reliably perform the holistic integration of molecular finding, patient phenotype, family history, and clinical context that a board-certified clinical geneticist or molecular pathologist brings to this step.