Preprint introduces GeneSIS framework to improve polygenic score transferability by modelling gene-by-sex interactions

A preprint on bioRxiv describes GeneSIS (GENE and Sex Interaction Score), a supervised statistical learning framework designed to improve the transferability of polygenic scores (PGS) across genetic ancestry groups by explicitly modelling gene-by-sex interaction effects at the variant level.

The authors analysed 406,659 individuals in the UK Biobank — including admixed individuals — and 1.3 million genetic variants to build predictive models for 99 complex traits. The core methodological innovation is the joint modelling of additive genetic effects and context-dependent (specifically sex-stratified) effects at single-variant resolution, rather than applying sex stratification post hoc at the score level. The framework is supervised in the sense that it is trained directly on individual-level data rather than relying solely on publicly available GWAS summary statistics.

Polygenic score transferability across ancestry groups is a widely recognised limitation of current PGS methodology, driven by differences in linkage disequilibrium patterns, allele frequencies, and gene–environment interactions across populations. The inclusion of sex as a context variable in GeneSIS addresses an additional axis of heterogeneity that is frequently treated as a covariate to adjust away rather than a modifier of genetic effect.

The preprint has not undergone peer review. Researchers in statistical genetics, genetic epidemiology, and the methodology of polygenic prediction — particularly those working on ancestry-diverse datasets — are the primary audience. Educators covering polygenic risk scores and their limitations may find the study a useful case study in model design.