Operationalizing language-based population stratification for widening access to precision genomics in Africa

Despite remarkable advancements in genomic technologies, individuals of predominant African-related genetic similarity remain significantly under-represented, accounting for only 2.4% of published genome-wide association studies. This disparity limits our understanding of human biology and hinders equitable translation of genomic advances into healthcare.

We exploited a quantitative framework using normalized Levenshtein distance (LDN) to analyse lexical similarity patterns across Kenya's ethnolinguistic landscape, comprising Bantu, Nilotic, and Cushitic language groups. We compared lexical distance matrices with available genetic population differentiation data and geographic proximity to evaluate their relative efficacy in predicting genetic relationships.

Lexical similarity analysis revealed distinct clustering patterns that closely mirror Kenya's ethnolinguistic diversity. Multidimensional scaling and hierarchical clustering clearly separated the three major language families and identified fine-scale relationships within each group. Importantly, lexical distance demonstrated stronger correlation with genetic differentiation [r = 0.91, CI (0.55-0.99)] than geographic proximity [r = 0.29, CI (0.29-0.53)], confirming language as a superior proxy for population genetic structure. Our analysis, demonstrate an objective basis for prioritizing populations in genomic studies.

This study establishes lexical similarity analysis as a powerful alternative approach for predicting genetic relationships among diverse African populations. By enabling strategic prioritization of representative populations for genomic sequencing initiatives, this approach offers a practical solution to address the critical under-representation of African genetic diversity in global databases, with potential applications across Africa's over 3,000 ethnic groups. This methodology provides a systematic, data-driven alternative to convenience sampling in regions where genetic data remains limited.