A linear-transformer hybrid for single-nucleotide polymorphism-based genotype-to-phenotype prediction in grapevine

Robust genotype-to-phenotype (G2P) prediction is essential for accelerating grapevine (Vitis spp.) breeding decisions and genetic gain. However, it remains challenging to measure complex traits under variable field conditions and across years. In this study, we propose a linear-Transformer (LiT)-G2P, an automated predictive framework that integrates additive genetic variance effects with Transformer-based nonlinear interactions using genome-wide single-nucleotide polymorphisms (SNPs) data. We evaluated LiT-G2P on a panel of diverse grape accessions, genotyped with SNP markers and measured for phenotypes across three consecutive years. Target phenotypic traits include leaf hair density and trichome density of grapevines. Across both single-year and cross-year testing scenarios, LiT-G2P consistently improves prediction performance compared with baseline models. For leaf hair density, LiT-G2P achieved the lowest prediction errors in both single-year and cross-year evaluations, with root mean square error (RMSE) values of 0.475 and 0.460 and ± 0.5 accuracies of 78.6% and 72.9%, respectively. For trichome density, LiT-G2P achieved RMSE values of 0.637 and 0.766 and corresponding accuracies of 62.6% and 58.2%. In addition, we extracted model-prioritized SNPs from attention weights and applied genotype-stratified analysis to provide interpretable candidate marker for downstream validation. These results support practical SNP-based predictive modeling for genomic selection and identify the remaining need for more detailed field data, larger multi-year grapevine panels, and independent grapevine validation before broader deployment.