We present Hi4GS, a hybrid feature selection (HFS) algorithm for selecting SNP subsets from high-dimensional genotypes to improve the prediction of genomic estimated breeding value (GEBV) under genomic selection (GS). Hi4GS combines feature importance weighting with quantity determining to construct a fused feature set from which it extracts an optimal feature subset for subsequent GS. In a study of wheat using four datasets covering 11 yield traits via large-scale GS models, the SNPs selected by Hi4GS increased the average predictive accuracy by over 82% than using all SNPs. Hi4GS was used to identify SNPs potentially affecting wheat yield, and SHAP-based interpretability was applied to explain the contributions of these SNPs and their potential interactions. Hi4GS can be used for assisting in improving the prediction accuracy of GS, wheat and other plants’ yield-associated SNPs identification, and target information for breeding chip development. The free R package Hi4GS is available at https://github.com/shgs19/Hi4GS.

Awns play an important role in seed dispersal and photosynthesis of spikes. Three major awn inhibitors (Hd, B1, and B2) are reported in wheat. However, the molecular mechanism underlying awnlessness remained unknown until recently. In this study, we identified two F₈ recombinant inbred lines (RILs) that were segregating for awn length. In order to identify the causal gene for awn length in the heterozygous inbred families (HIFs), SNPs were called from RNA sequencing (RNA-Seq) data for HIF-derived progenies with long and short awns. SNPs between long and short awn plants were evenly distributed on chromosomes (chr) other than chromosome 5A. SNPs on chr 5A were clustered in a region distal 688 Mb on the long arm, where inhibitor B1 was located. This suggested that B1 was the causal segregating locus. We precisely mapped B1 to ~1 Mb region using two HIF-derived families. Considering that the lines segregated for long, intermediate and short awn phenotypes we speculated that B1 should have a dosage effect on awn length. Two differentially expressed genes (DEGs) located in the candidate region were regarded as candidate genes for B1, because the molecular expression pattern was consistent with the phenotype. HIFs with long and short awns showed no difference on grain yield and other agronomic traits.