Abstract

Iron deficiency anemia disproportionately affects South Asian populations, yet population-specific genetic risk models for this group remain largely absent, as most existing polygenic risk score (PRS) frameworks were derived from European cohorts with limited transferability across ancestries. This study examined the feasibility of classifying genetic risk profiles for iron deficiency anemia in South Asian individuals using simulated PRS and support vector machine (SVM) classification. Genetic data from 489 individuals across five South Asian subpopulations were integrated with iron-related GWAS summary statistics, and each individual was assigned a PRS representing cumulative genetic predisposition. Disease labels were simulated through a Liability Threshold Model (LTM) that assigned case and control status probabilistically. Two SVM classifiers were compared: one using the aggregated PRS as input, and one using individual genetic markers under Leave-One-Chromosome-Out (LOCO) cross-validation to prevent data leakage. The PRS-based classifier achieved AUC-ROC=0.701 and recall=0.727, while the marker-based classifier produced near-random performance with AUC-ROC=0.509. The AUC gap of 0.192 between configurations was the primary finding, demonstrating that PRS aggregation is a necessary preprocessing step where sparse individual markers carry insufficient discriminative signal at this sample size. The study contributed a reproducible, South Asian-focused pipeline built from publicly available data, extensible to real clinical data as genomic resources for this population expand.

Keywords