Sorghum grain mold is a devastating fungal complex whose genetic dissection is frequently hindered by polygenic architecture and inconsistent marker signals. To resolve this complexity, we developed a unified GWAS framework that integrates standard single-nucleotide polymorphisms (SNPs) with a local haplotype-window marker layer (“Haplo-kmer;” k-SNP windows encoded from the GBS SNP matrix, not read-derived k-mer counts), prioritizing genomic intervals where evidence converges across multiple traits and methods. Leveraging the Sorghum Association Panel (SAP) and six distinct phenotype indices, we identified a convergent core interval under an operational multi-trait, multi-method convergence framework (p \(\le 5\) × 10–4) on Chr5 (60.25–60.50 Mb). This interval was supported by multi-trait convergence and rank-preserving permutation analyses, representing a convergent hub supported by three distinct resistance-related traits and reinforced by haplotype-based association signals ( \(\Sigma -{\text{log}}_{10}\left(p\right)=32.43\) ). Decomposition of the signal revealed that while SNPs captured isolated hits for Alternaria (A) and Curvularia (C) indices, the mixed-inoculum (M) index was supported by both SNP and Haplo-kmer markers, alongside strong contributions from A and C indices. We validated the biological relevance of this locus through unsupervised haplotype clustering of 54 regional SNPs, which successfully stratified sorghum lines into distinct phenotypic classes (Kruskal–Wallis p = 0.0029). Furthermore, a genome-wide placebo test demonstrated that this haplotype–phenotype stratification shows moderate specificity against the random genomic background (empirical p = 0.165). Collectively, these findings delineate a robust candidate interval for marker-assisted tracking on Chr5 and demonstrate how haplotype-aware strategies can recover fragmented association evidence in a polygenic pathosystem.
