A machine learning approach to genome-wide association mapping of disease resistance and geographic origin in sorghum.

Sorghum, the fifth most important cereal crop globally, faces persistent threats from fungal diseases that limit productivity and resilience. To investigate the genetic basis of disease resistance and geographic adaptation, we applied a machine learning-enabled genome-wide association study (GWAS) to a panel of 377 genetically diverse sorghum accessions, incorporating nearly 300,000 SNP markers and phenotypic evaluations for resistance to anthracnose, head smut, and downy mildew. While disease resistance phenotypes did not cluster strictly by geographic origin, SNP-based analyses revealed significant genetic differentiation among accessions from different regions, particularly involving a genetically distinct group from Senegal. Bootstrap Forest models highlighted candidate SNPs predictive of geographic origin, most notably on Chromosome 10, near genes encoding transcription factors (e.g., bHLH, EREBP-like) and DUF6598-domain proteins with potential roles in plant defense. For disease resistance, top-ranked SNPs were located near genes implicated in canonical immune pathways, including zinc-binding proteins (anthracnose), NB-ARC and LRR-containing proteins (head smut), and F-box proteins (downy mildew). Although exploratory in nature, these findings suggest that local adaptation to pathogen pressure may have shaped sorghum’s genomic landscape. The identified candidate genes and associated SNPs help prioritize targets for marker-assisted selection and follow-up functional validation, contributing to the development of sorghum varieties with enhanced resistance and adaptability.