Preprint proposes conditional SNP-heritability framework for ancestrally diverse datasets

A preprint posted to bioRxiv introduces a framework that distinguishes between marginal SNP-heritability — a single estimate computed across an entire genetically diverse dataset — and conditional SNP-heritability, which is estimated separately within ancestral subpopulations.

SNP-heritability is the proportion of phenotypic variance explained by the genetic variants assayed in a genome-wide association study (GWAS). Standard methods return one figure per trait per dataset. The authors argue that when a dataset pools individuals from multiple ancestral backgrounds — each with potentially different environmental exposures, allele frequencies, and linkage disequilibrium structures — a single marginal estimate conflates genuine biological heterogeneity with statistical artefact.

The conditional framework allows researchers to ask whether a trait's heritability differs between subpopulations, which has implications for understanding gene–environment interactions and for the transferability of polygenic scores across populations. The paper presents theoretical derivations and applies the method to empirical datasets.

This work is relevant to researchers developing or applying GWAS methodology in diverse cohorts, and to statistical geneticists working on polygenic score portability. It connects to a broader ongoing conversation about the assumptions embedded in heritability estimation and the importance of ancestral diversity in genomic research.

This study is a preprint and has not yet been peer-reviewed.