maasMDS: Multi-array ancestry-specific multidimensional scaling for fine-scale structure in admixed genomes

Motivation Fine-scale population structure in admixed cohorts is often obscured by genome-wide dimensionality reduction, which averages signals from different genetic ancestry sources present in the same individual, and by batch effects when datasets are merged across heterogeneous genotyping arrays with sparse SNP overlap. Results We present multi-array ancestry-specific multidimensional scaling (maasMDS), a framework that combines local-ancestry masking with distance calibration to integrate samples genotyped on different arrays while resolving structure within a chosen ancestry component. By calibrating ancestry-specific pairwise distances on shared observed loci before embedding, maasMDS mitigates array-driven clustering without discarding large numbers of variants and more faithfully resolves fine-scale relationships than post hoc Procrustes alignment or standard imputation-based harmonization. Across analyses of Indigenous American cohorts, maasMDS recovers sub-continental origins of minority European ancestry segments consistent with regional colonial histories and reveals geographic clines within Indigenous populations of Mexico and Central America that are obscured without ancestry restriction or multi-array correction. Availability and Implementation maasMDS is implemented in the snputils package with both a Python API and CLI. Source code is available at https://github.com/AI-sandbox/snputils.