Identification of Genomic Regions for Partial Resistance to Soybean Rust Under Field Conditions Using FarmCPU and Machine Learning Approaches

Abstract

Soybean rust caused by the fungus Phakopsora pachyrhizi threatens global soybean production, causing yield losses of up to 80%. Race-specific Rpp genes provide short-term resistance due to pathogen variability, whereas partial resistance (PR) offers durable, broad-spectrum protection, though its genetic basis remains unclear. This study aimed to identify genomic regions and candidate genes underlying PR using the Fixed and Random Model Circulating Probability Unification (FarmCPU) genome-wide association study (GWAS) and machine learning (ML) methods, Random Forest (RF) and Support Vector Regression (SVR). A panel of 312 soybean accessions was evaluated under natural infection across six Ugandan environments. Rust index (RI), derived from rust severity and sporulation level, was used to estimate heritability (H2) and rank genotypes through Best Linear Unbiased Predictions (BLUPs), while Best Linear Unbiased Estimators (BLUEs) supported GWAS input. After quality control, 8272 SNPs were analyzed within a ±60 kb linkage disequilibrium (LD) window. Multi-environmental Analysis (MEA) of RI showed significant genetic effects (p < 0.01); H2 = 0.57–0.68. Sixty-one loci were detected: six by FarmCPU, 15 by RF, and 41 by SVR. Key genes included Glyma.01G128100 (a WRKY transcription factor) and Glyma. 13G228000, receptor-like kinase) and Glyma.20G173100 (WD40-domain regulator). Integrating ML with GWAS improved locus detection, confirming the polygenic nature of PR and supporting the use of genomic selection and locus pyramiding for durable rust resistance.