Integrated single marker scanning and sparse Bayesian learning improves performance of detection for GWAS

Single-locus and multi-locus models provide a statistical framework for conducting genome-wide association studies (GWAS). However, mainstream GWAS algorithms face challenges such as low detection power, high false positive rates, and slow computational efficiency as the dimensionality of genomic data continues to grow. We introduce KinLmSBL, a two-stage approach that integrates single-locus scanning with polygenic background control and multi-locus sparse Bayesian learning. We validate the approach through a series of simulations and demonstrate that KinLmSBL outperforms four other single- and multi-locus approaches in detecting low heritability variants, controlling the false positive rate, and increasing efficiency. We apply KinLmSBL to maize, rice, and human datasets, further investigating its performance in identifying more known genes with lower computational cost. Overall, KinLmSBL provides an efficient GWAS tool for mining genes in high-dimensional biological data.