Spatially-integrated Ancestry Polygenic Risk Mapping for Diabetes: A Geographic-adjusted Genetic Scoring Framework

Background: Conventional polygenic risk scores (PRSs) often fail to capture geographic and ancestry-specific variation, limiting their predictive accuracy in admixed and globally diverse populations. This study aims to develop a spatially-aware PRS (SA-PRS) framework that integrates ancestry-specific genetic contributions with geographic admixture patterns to enhance diabetes risk prediction. Methods: The SA-PRS framework incorporates ancestry-aware variant weighting with spatial modelling approaches. It utilises admixture-spatial decomposition to estimate individual ancestry proportions and geographic distribution through a transformer-based variational autoencoder. Hierarchical Bayesian spatial weighting is applied to model region-specific allele frequency distributions, while a three-dimensional spatial risk weighting tensor to dynamically adjust variant effects across geographic contexts. In addition, an attention-based fusion network integrates genetic risk with clinical risk factors to ensure compatibility with conventional models. Results: The proposed model demonstrated a statistically significant improvement in diabetes risk stratification in admixed populations ( P < 0.05), with enhanced predictive performance compared to conventional PRS models. It identified spatial-genetic interactions, including migration-driven allele frequency gradients and regional epistatic effects, highlighting the geographic heterogeneity of diabetes susceptibility. Conclusion: The SA-PRS framework provides an interpretable and scalable approach for precision medicine, although further validation in larger multiethical studies is required to confirm generalisability.