Calibration and transferability
Performance, ancestry, reference population, and local context determine whether a relative position can be interpreted responsibly.
Scientific foundation
Scientific rigour comes first. We base analyses and reporting on peer-reviewed publications, current professional guidelines, validated reference data, and proven laboratory and analytical techniques. Conclusions never go beyond what the data, evidence, and intended use support.
From polygenic score to clinical context
Polygenic scores are probabilistic instruments. They estimate inherited predisposition relative to a reference population; they do not diagnose disease or determine an individual outcome.
Starling therefore starts with intended use, source-data quality, model provenance, the reference population, and the conclusion that can responsibly be reported. A score becomes useful only when limitations, evidence, version, and action context remain visible to the professionals responsible for interpretation and implementation.
Calibration, ancestry, transferability, and comparison with established clinical predictors are assessed before implementation. A statistically strong model can still be unsuitable for a local population or care pathway without appropriate validation and clear reporting boundaries.
Example · PRS interpretation
One modality: relative position, not diagnosis
A polygenic score places an individual within a reference distribution. Its clinical meaning still depends on calibration, ancestry, and the setting in which the result will be used.
What supports a responsible PRS
Model performance is assessed through discrimination, calibration, external validation, ancestry context, and comparison with predictors already used in care. Published association alone is not enough to justify implementation.
The intended use determines which evidence is relevant: population stratification, prevention, pathway triage, or research may require different thresholds, reference distributions, and local validation. Reporting must make those boundaries explicit.
Representative knowledge and validation resources
Population and validation
UK Biobank, All of Us, ancestry-diverse validation cohorts
Score and model resources
PGS Catalog, GWAS Catalog, published model specifications
Reference and imputation
1000 Genomes, TOPMed, fit-for-use reference panels
Literature and domain sources
PubMed, GWAS Catalog, domain-specific cohorts
Interpretation principles
Polygenic scores support probabilistic stratification; they do not diagnose disease or predict an individual outcome with certainty. They should be interpreted alongside age, family history, biomarkers, clinical measurements, and the intended care pathway, with explicit evidence and validation boundaries.
Independent validation matters
External cohorts and local data show whether model performance remains reliable outside the original development environment.
Model provenance stays visible
Publications, weights, genome build, reference distribution, pipeline version, and quality criteria remain traceable in the result.
Translation stays cautious
The scientific objective is not to overclaim maturity, but to identify where current evidence is strong enough to justify structured validation and careful implementation work.
Ecosystem
Technology and service relationships
We combine genotyping, laboratory, and PRS analytics capabilities according to intended use, sample type, scale, and implementation setting.
Research and knowledge resources
Scientific work may draw on appropriately governed cohorts, evidence resources, and clinical knowledge bases.
UK Biobank · All of Us Research Program · PGS CatalogIn practice, every workflow records which publications, guidelines, reference data, pipeline versions, and quality criteria support the reported conclusion—and which limitations remain.
Looking for a deeper literature review across disease areas and quantitative traits? Read our Research & Background overview.
Discuss a study or rollout?
Use the contact form and we will send your message to the right part of the team.
Contact