LiT-G2P: Hybrid Model Improves Grapevine Trait Prediction
Researchers propose LiT-G2P, a linear-Transformer hybrid for genotype-to-phenotype prediction in grapevine. The model integrates additive genetic variance with Transformer-based nonlinear interactions using genome-wide SNP data. Tested on diverse grape accessions across two years, LiT-G2P outperforms baseline models in predicting leaf hair and trichome density, achieving lowest error for hair density.
Key facts
- LiT-G2P stands for Linear-Transformer Genotype-to-Phenotype.
- It integrates additive genetic variance effects with Transformer-based nonlinear interactions.
- The model uses genome-wide single-nucleotide polymorphisms (SNPs) data.
- Evaluated on a panel of diverse grape accessions across two consecutive years.
- Target traits include leaf hair density and trichome density.
- LiT-G2P consistently improves prediction performance in single-year and cross-year tests.
- For hair density, LiT-G2P achieves the lowest error.
- The approach aims to accelerate breeding decisions and genetic gain.
Entities
—