SPLENDID incorporates continuous genetic ancestry in biobank-scale data to improve polygenic risk prediction across diverse populations.

Penalized regression on biobank-scale individual-level data, modelling genetic ancestry as a continuum

SPLENDID is a penalized regression framework for biobank-scale individual-level data that models genetic ancestry as a continuum, without labelling individuals by ancestry group. It produces a single prediction model, removing the practical requirement to assign every person to a pre-specified ancestry category — a step poorly suited to clinical decision-making and impossible for many admixed individuals. In extensive simulations and in the All of Us Research Program (n = 224,364) and UK Biobank (n = 340,140), SPLENDID significantly improved prediction accuracy over existing methods, particularly for non-European and admixed ancestries.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

The problem with polygenic risk scores is not only transferability: applying a model requires slotting the patient into an ancestry category, which no clinical encounter actually does. A single continuous model is the right methodological answer, validated here in two biobanks of over 200,000 participants each. What is missing is the one thing that would decide clinical use: reclassification metrics against a real decision — screening threshold, age at which surveillance starts — rather than a gain in prediction accuracy.

Analysis by Dr Thibaut Benquey