Tertiary Analysis in NGS
From Variant List to Clinical Diagnosis

The NGS pipeline is divided into three stages, each building on the last. The boundary that matters most for clinical labs is between secondary and tertiary, where the work shifts from computational to interpretive.

Secondary analysis has been largely solved. The tools are validated, the pipelines documented, the compute requirements understood. Tertiary analysis has not been solved in the same way, for four structural reasons.

The labs with the highest diagnostic yields and shortest turnaround times invested in platforms that automate the repetitive, evidence-gathering portions of interpretation, freeing clinical geneticists to focus on genuinely ambiguous cases.

Annotation adds biological context to each raw variant call. A position and base change conveys nothing clinically useful on its own. Annotation transforms it into a clinically evaluable entity by answering: what gene does this affect, what does the change do to the protein, has it been seen before, how common is it in the population, and is it already classified in any clinical database?

Essential annotation sources

After annotation the goal is reduction: remove the vast majority of variants that are clearly not responsible for the patient's disease, while retaining every variant that might be. The tension between sensitivity (do not miss the answer) and specificity (do not bury it in noise) defines filter chain design.

Filtering removes what is clearly not the answer. Prioritization ranks what remains by how likely each variant is to explain the patient's disease.

Phenotype-driven gene ranking

The patient's clinical presentation is the most powerful prior for prioritization. HPO (Human Phenotype Ontology) terms, a standardized vocabulary for clinical features, are used to rank genes by their relevance to the patient's phenotype. A missense variant in a gene with established association to the patient's specific combination of features should surface to the top of the review queue. The same variant in a gene with no phenotypic connection should be deprioritized. Tools like PhoRank (integrated in VarSeq) compute phenotypic similarity scores between the patient's HPO terms and each gene's known phenotypic spectrum.

Gene-disease validity

Not all reported gene-disease relationships are equally supported. A variant in a gene with Definitive ClinGen classification deserves more weight than a variant in a gene with Limited or Disputed classification. Many gene-disease relationships in the literature were established before modern statistical standards. Some have been refuted or weakened.

Variant-level evidence aggregation

Before formal classification, the interpretation platform should surface all available evidence for each candidate variant: ClinVar submissions and star ratings, published case reports from HGMD or Mastermind, gnomAD frequency, in silico predictions, conservation, and the lab's own knowledgebase. This is where the quality of the tertiary analysis platform has its largest impact on turnaround time. A platform that automatically surfaces and organizes this evidence lets a clinical geneticist evaluate a candidate in minutes rather than hours.

Classification assigns a clinical pathogenicity tier to each candidate variant using standardized frameworks. Germline uses one framework. Somatic uses another. Labs running both need to handle both.

Germline: ACMG/AMP guidelines

The 2015 ACMG/AMP guidelines (Richards et al., Genetics in Medicine) define a five-tier classification system: Pathogenic, Likely Pathogenic, Variant of Uncertain Significance, Likely Benign, Benign. Classification is determined by applying 28 criteria organized by evidence type and strength, from PVS1 (null variant in a gene where loss-of-function is a known mechanism) to BP7 (silent variant with no predicted splice impact). Each criterion is weighted Very Strong, Strong, Moderate, or Supporting. The combination of met criteria determines the final tier. ClinGen Expert Panels publish gene-specific rule refinements for many clinically important genes.

Somatic: AMP/ASCO/CAP guidelines

Somatic classification (Li et al., 2017) follows a different framework. Unlike germline (focused on pathogenicity), somatic focuses on actionability: relevance to treatment, prognosis, or trial eligibility. Four tiers:

• Tier I, Strong clinical significance: FDA-approved therapies or standard of care support use for treatment decisions in this tumor type.
• Tier II, Potential clinical significance: Clinical relevance from trials, case reports, or evidence in the same or different tumor type.
• Tier III, Unknown clinical significance: Not established as clinically relevant. May be a passenger mutation.
• Tier IV, Benign or likely benign: Common variant with no established role in cancer.

The VUS problem

A VUS is the most common outcome of diagnostic sequencing for rare disease patients. Rates of 30 to 50% are typical in WES, not because the technology is failing but because current knowledge is genuinely insufficient to classify many variants definitively.

A VUS is not a dead end. It is the beginning of an evidence-accumulation process. Functional studies, family segregation, continued absence from gnomAD as the dataset grows, new ClinVar submissions, and lab-internal observations all contribute over time. Reclassification rates of 10 to 30% over 3 to 5 years are documented across large lab cohorts. Labs have a clinical obligation to monitor VUS reclassification and notify patients when status changes. That requires a variant knowledgebase that tracks every VUS ever reported and alerts staff when its external classification shifts.

The clinical report is the final output of tertiary analysis. The document that reaches the ordering physician, enters the medical record, and directly influences patient care. Its content, format, and accuracy are subject to CAP/CLIA standards.

Required report elements

• Patient and specimen identification
• Test methodology: sequencing platform, capture kit, reference genome build, bioinformatics pipeline version, variant caller(s)
• Analytical performance metrics: mean coverage, percentage of target bases at minimum depth
• Reported variants with HGVS nomenclature (genomic, coding, protein)
• Variant classification with a supporting evidence summary
• Clinical interpretation: how the identified variant(s) explain or fail to explain the phenotype
• Secondary findings policy: whether ACMG-recommended secondary findings were evaluated and reported
• Limitations: known coverage gaps, variants not detectable by this assay, VUS uncertainty
• Ordering provider information and laboratory director signature

Secondary findings

The ACMG recommends labs report secondary findings: pathogenic or likely pathogenic variants in 81 medically actionable genes (SF v3.2), when identified incidentally during exome or genome sequencing, regardless of the primary indication. Implementation is not universal. Patients should be counseled and offered the option before testing. Secondary findings reporting requires a clear lab policy, documented patient consent, and a clinical process to communicate findings to appropriate specialists.

EHR integration

Reports should be delivered through secure, trackable channels: a laboratory information system (LIS) or direct EHR integration. PDF reports via secure portal are standard. HL7 FHIR-based structured reporting for EHR integration is increasingly available and enables downstream clinical decision support.

Research-grade tools, even excellent and widely used ones, are not appropriate for clinical use without substantial additional validation and infrastructure. Seven criteria define a clinical-grade tertiary analysis platform.