Clinical evidence
Different kinds of evidence answer different questions.
A computational prediction, a laboratory experiment, a retrospective observation and proof of clinical utility are not interchangeable. This page keeps them apart, capability by capability.
Evidence taxonomy
Seven kinds of evidence
Computational prediction
An estimate produced by an algorithm or model from available data.
Research evidence
Published or curated scientific findings, of varying strength and relevance.
Functional / experimental
Observations from laboratory or model-organism experiments, such as AVATAR's fruit-fly screening.
Retrospective observation
Patterns seen when past cases are analysed; useful but prone to bias.
Analytical validity
Does the test accurately measure what it claims to measure?
Clinical validity
Is the result reliably associated with the clinical condition or outcome of interest?
Clinical utility
Does using the result improve decisions or outcomes for patients?
By capability
The evidence behind every capability
Each capability, with the kind of evidence behind it.
| Capability | Evidence type | What supports it |
|---|---|---|
| Twin testing pathways (Blood 600, Exome, Liquid 523, Tissue 335) | Laboratory assays | Performed in CAP- and NABL-accredited laboratories, with assay validation and external proficiency testing. |
| Twin analytical layers (1–21) | AI-assisted computational analysis | Validated pipelines, curated knowledge bases and versioned methods across 21 proprietary layers. |
| Twin report and clinician review | Clinician-led interpretation | Five report components, signed by the treating oncologist and reviewed by the KPCIRC Molecular Tumour Board. |
| Clinical-trial intelligence (TrialGraph) | Evidence matching | Matching against clinical-trial registries on biomarkers, cancer type and eligibility criteria. |
| OnKommon Interpreter | Computational interpretation | Driver discovery across tumour cohorts, mutation-level driver prediction and a report-trained language model — run through the six proprietary core layers. |
| AVATAR — Discovery Process | Functional / experimental (Drosophila) | Patient-specific fruit-fly models, engineered from the tumour's drivers and screened against candidate drugs and combinations. |