A Machine Learning Approach to Genome-Wide Association Mapping of Disease Resistance and Geographic Origin in Sorghum

Background Sorghum, often considered the fifth most important cereal crop globally, faces significant production constraints caused by various fungal diseases. Understanding the genetic basis of disease resistance and adaptation to geographic origin is crucial for developing improved varieties. This study investigates these aspects in a diverse panel of 377 sorghum accessions using a machine learning-enabled genome-wide association study (GWAS). Results The study analyzed a panel of 377 sorghum accessions, including a mini core collection and additional accessions from Senegal. Phenotypic evaluation for resistance to anthracnose, head smut, and downy mildew was conducted on the mini core collection. Genotypic data comprising nearly 300,000 SNP markers were used for GWAS with Bootstrap Forest models. While phenotypic clustering based on disease resistance did not directly correlate with geographic origin, significant genetic differentiation was observed based on geographic origin. Machine learning-driven GWAS identified SNPs associated with geographic origin, particularly on chromosome 10, with candidate genes including transcription factors. SNPs near genes with known or predicted roles in plant defense were identified for disease resistance, such as zinc-binding proteins for anthracnose and LRR- and NB-ARC-containing proteins for head smut. Conclusions This research provides insights into the complex genetic architecture of disease resistance and geographic adaptation in sorghum. In addition to previously known resistant genes through traditional GWAS, the identified candidate genes and associated SNPs offer valuable resources for enhancing disease resistance in sorghum breeding programs through marker-assisted selection and other advanced breeding techniques.