03 · The OnKommon Interpreter

OnKommon Interpreter

Every genetic report hides its drivers.

A tumour's sequencing report can list hundreds of variants. Only a handful drive the cancer. The OnKommon Interpreter finds them — and shows why.

See the flow

01 · Driver knowledge

Built on the map of cancer drivers.

Driver knowledge drawn from tens of thousands of sequenced tumours across dozens of cancer types — with IntOGen's compendium of cancer driver genes at its core.

02 · Mutation-level prediction

Every mutation, judged in context.

BoostDM models each gene in each tumour type, telling drivers from passengers down to the single amino acid. OncodriveMUT extends the read; ClinVar and gnomAD set inherited and common variants apart.

03 · Language intelligence

A language model that reads reports.

A large language model, fine-tuned on clinical-genomic reports, reads the report the way a specialist does — context, biomarkers, wording — and reasons across it.

04 · One engine

Two intelligences. One interpreter.

Driver science and language intelligence fuse into a single engine — the OnKommon Interpreter.

05 · Six proprietary layers

Then through six proprietary layers.

CloneTrace reconstructs the tumour's evolution. SourceGuard separates what the tumour acquired from what the patient inherited.

06 · Signatures and phenotype

The tumour's fingerprint.

SigPrint reads the mutational signatures — the causes written into the genome. PhenoMap connects genotype to how the cells behave.

07 · Immunity and action

From immune landscape to action.

ImmuneLens maps the immune biomarkers. AccessMatch connects every actionable alteration to the therapies and evidence that target it.

08 · The interpretation

Drivers, biomarkers, evidence — ranked.

Every finding arrives with its evidence strength and therapeutic associations, and anything uncertain is flagged for expert review.

09 · Then AVATAR

Then the biology is tested.

With the drivers known, OnKommon AVATAR carries them into a living fruit-fly model — a biological investigational plan.

Continue to AVATAR

The Interpreter

Hundredsofvariants.Ahandfulofdrivers.TheInterpreterfindsthem.

The flow

From genetic report to living model

Every report follows the same path — interpretation first, then the six proprietary layers, then AVATAR.

01Genetic reportThe sequencing variants, exactly as the laboratory reports them.
02Driver knowledgeIntOGen · BoostDM · OncodriveMUT · ClinVar · gnomAD
03Language modelFine-tuned to read and reason over clinical-genomic reports.
OnKommon InterpreterTwo intelligences, fused into one engine.
05Six proprietary layersCloneTrace · SourceGuard · SigPrint · PhenoMap · ImmuneLens · AccessMatch
06OnKommon AVATARThe drivers, carried into a living model.

The knowledge inside

The science of cancer drivers, in one engine

Peer-reviewed driver science and report-trained language intelligence — each doing what it does best.

Driver discovery
IntOGen
A compendium of cancer driver genes, discovered by running multiple driver-detection methods across tens of thousands of sequenced tumours and dozens of cancer types.
Mutation-level driver prediction
BoostDM
Machine-learning models built per gene and per tumour type, telling driver mutations from passengers down to the single amino acid.
Rule-based driver assessment
OncodriveMUT
Extends the read to mutations in cancer genes without a dedicated model, using features such as mutation consequence, clustering and conservation.
Clinical significance
ClinVar
The reference archive of variant–condition relationships, used to recognise inherited variants of known clinical significance.
Population frequency
gnomAD
Allele frequencies across large populations, used to set common germline variation apart from tumour-acquired change.
Report intelligence
Language model
A large language model fine-tuned on clinical-genomic reports, reading context, biomarkers and wording the way a specialist does.

What it returns

An interpretation the oncologist can act on

01
Driver alterations
The mutations that drive this tumour, ranked — separated from the passengers around them.
02
Biomarkers
TMB, MSI, HRD and other genomic biomarkers, read in the context of the cancer type.
03
Evidence strength
Each finding graded by the strength and source of the evidence behind it.
04
Therapeutic associations
Sensitivity and resistance, connected to therapies and clinical trials.
05
Inherited findings
Germline variants of clinical significance, set apart for confirmation and counselling.
06
Uncertain findings
Variants of uncertain significance, flagged clearly for expert review.

Where it runs

Inside every OnKommon journey

Questions

The Interpreter, explained

What is the OnKommon Interpreter?

The OnKommon Interpreter is OnKommon's interpretation engine for tumour genetic reports. It combines cancer-driver knowledge with a language model fine-tuned on clinical-genomic reports, then runs every finding through six proprietary core layers.

Which knowledge resources does it draw on?

Driver discovery from IntOGen, mutation-level driver prediction from BoostDM, rule-based driver assessment from OncodriveMUT, clinical significance from ClinVar and population frequencies from gnomAD.

What does the language model add?

It reads the genetic report the way a specialist does — the cancer type, the biomarkers, the laboratory's wording — so that each variant is interpreted in the context of the whole report.

How does the Interpreter relate to Twin and AVATAR?

The Interpreter feeds the six shared core layers — CloneTrace, SourceGuard, SigPrint, PhenoMap, ImmuneLens and AccessMatch. In AVATAR, the drivers it identifies guide the design of the fruit-fly model and the candidate-drug screen.

Who signs the interpretation?

The treating oncologist reviews and signs every report, and KPCIRC's Molecular Tumour Board reviews the findings.

See the Interpreter read a genetic report.

A professional demonstration for oncologists, hospitals, laboratories and research teams.

Explore AVATAR Request a demonstration