Somatic Variant Interpretation in Cancer Genomics: Beyond the Variant Call

Every one of these frameworks runs on curated evidence, and in cancer that evidence lives in a few knowledgebases most labs come to rely on. 

• • • OncoKB, maintained by Memorial Sloan Kettering, grades variant-drug relationships on its own levels of evidence and in 2021 became the first somatic database to earn partial FDA recognition (6, 7). 
    • CIViC takes an open, community-curated approach, with every interpretation traceable back to the literature and free for anyone to use (8). 
    • CKB, the Cancer Knowledgebase, works along the same lines and pairs each variant or molecular profile with the therapies, clinical evidence, and open trials attached to it, grades that evidence on the AMP/ASCO/CAP tiers, and traces each entry back to its guideline (NCCN, ESMO) or primary-literature source (9). 
    • COSMIC plays a different role. It is primarily a catalogue of which mutations have been seen across cancers, along with their frequencies and associated mutational signatures (10).

Oncogenicity vs. Actionability

Oncogenicity is largely a property of the variant. An alteration that disables a tumor suppressor or switches on an oncogene does so wherever it occurs, so its oncogenicity stays the same from one tumor to the next.  Actionability depends on context: the tumor type, the drugs that exist, and the evidence behind them. A variant can therefore hold a fixed oncogenicity and still move between clinical tiers depending on where it shows up. 

BRAF V600E shows this plainly. It is an oncogenic driver in both melanoma and colorectal cancer, and that classification does not change between them. What changes is its clinical tier which depends on the drug and the tumor type. In melanoma, V600E paired with a BRAF inhibitor is a Tier I, ready-to-use target, and the drug given alone produces strong responses (11). In colorectal cancer the same variant does not reach that tier for the same drug: a BRAF inhibitor alone works poorly, because blocking BRAF triggers feedback activation through EGFR that keeps the cell growing (12). The variant only becomes Tier I in colorectal cancer when paired with a different strategy, a BRAF inhibitor together with an anti-EGFR antibody (13). One variant, one oncogenicity, but the same drug earns a different tier in each tumor. 

This is also why a clinical tier is a snapshot, not a fixed label. Oncogenicity reflects the biology of the variant and stays put; a tier reflects the evidence available at the time. That evidence grows, so the same variant can become more actionable over time without its biology changing at all.

What the Allele Fraction Tells You

The variant allele fraction is the share of sequencing reads that carry the change. In germline testing it barely moves: about half the reads for a heterozygous variant, nearly all for a homozygous one, so it confirms little you don’t already know. In a tumor it does real work. A high fraction points to a variant present in most of the cancer’s cells, and likely an early event; a low one points to a smaller subpopulation, and a later event (14). 

Allele fraction is only as clean as the sample it comes from. It reads as a proportion of tumor cells only when the sample is nearly pure tumor at normal copy number, which it often is not. Low tumor purity drags every fraction down at once, so a clonal driver can look subclonal. Copy-number changes distort it locally too: an amplification or a loss of heterozygosity shifts the fraction of any variant sitting on the affected segment. 

An allele fraction is read against the tumor’s purity and local copy number, never on its own. 

The reads underneath the fraction have to be trustworthy as well. At low fractions a real variant becomes hard to separate from a PCR or a sequencing error, and this is where unique molecular identifiers (UMIs) earn their place. Each original DNA fragment is tagged before amplification, so duplicate reads collapse back to the single molecule they came from (15). UMIs therefore make confident calling possible at very low fractions which is typical for circulating tumor DNA and liquid biopsies (16). 

How the sample was sequenced matters just as much. With a matched normal, the patient’s own germline variants can be subtracted directly, leaving a cleaner set of genuinely somatic calls. A tumor-only run has no such comparison, so it relies on population-frequency databases to infer which alterations are inherited, and some germline variants inevitably slip through disguised as somatic.

Sample-Level Signals: TMB and MSI

Alongside the question of whether a specific variant is oncogenic and actionable, some of the most consequential decisions in oncology rest on a property of the tumor as a whole. Tumor mutational burden (TMB) and microsatellite instability (MSI) are two such genome-scale biomarkers. Both are read out of the same sequencing data used for variant calling, and both tend to point toward immunotherapy rather than a targeted drug. 

Tumor mutational burden (TMB) is, in essence, a count: the number of somatic mutations per megabase of tumor DNA (17). The biological logic is that a heavily mutated tumor produces more abnormal proteins, making it more visible to the immune system and more likely to respond to immune checkpoint inhibitors (18). This is the reasoning behind pembrolizumab’s approval for TMB-high (≥10 mutations/Mb) solid tumors. Thresholds are assay-dependent and the predictive value varies by tumor type. 

Microsatellite instability (MSI) is a signature of a broken repair system. When a tumor’s DNA mismatch repair (MMR) machinery fails, short repeated motifs called microsatellites accumulate insertions and deletions, and the tumor reads as MSI-high. A tumor like this is effectively mismatch-repair deficient, and that status carries real weight. It was the basis of the first cancer treatment the FDA cleared on molecular grounds alone, regardless of where the tumor arose (19). High TMB and MSI-high often show up in the same tumor, but they are not the same measurement, and either one on its own can open the door to a checkpoint inhibitor.

SEQ Platform: From Variant Call to Clinical Decision

SEQ Platform is built around these distinctions. It separates somatic from germline calls, supports both classification and actionability, and reports the sample-level biomarkers alongside them. SEQ Platform earns its place by keeping the right context beside each of those calls instead of flattening them into a single verdict. 

SEQ Platform draws somatic clinical evidence and trial matching from CKB and its supporting literature from Mastermind (20). Detected biomarkers are matched against CKB during analysis, and each match carries its AMP/ASCO/CAP tier and level, the therapy and tumor type it applies to, and the trials that require it. Because a variant’s tier depends on the tumor type, the cancer term can be updated in the interface, and the matched evidence, its tiers, and evidence levels update automatically to reflect the new context. SEQ then shows the strongest matched tier for each variant, so the analyst starts from an evidence-based tier rather than grading each one by hand. Because evidence is sometimes written for an exact variant and sometimes for a codon, an exon, or a whole gene, SEQ also lets the analyst set how specific a match has to be to count, so the patient’s precise change and the broader evidence around it stay visible together. 

The judgments are kept distinct but shown side by side. For a single variant, the analyst sees the ACMG classification, the ClinVar entry, the matched evidence and its tier, the allele fraction and depth, and the population frequencies in one row and also has the “My Verdict” option to record or override the call. The classification and the actionability sit next to each other rather than being collapsed into one number. 

The sample-level context lives in the same interface. The TMB and MSI values discussed above have their own card; the tumor-only or matched-normal choice selects a different analysis pipeline; and UMI-aware pipelines are available for solid-tumor and ctDNA work, so a low allele fraction can be trusted for what it is. Targeted callers for FLT3-ITDs, on-target rates for gene panels, and quality-control checks each have a place in the same view. 

From there the analyst flags what matters into a Shortlist and turns it into a customizable clinical report. SEQ runs in the cloud and is CE-IVD certified and GDPR compliant. What matters for the work described here, though, is simpler: the separate judgments stay separate, each sits beside the context it has to be read against, and the analyst stays the one who decides.