A1200
Title: Elucidating the temporal dimension of the genotype by environment interaction effect with transformers
Authors: Patrick McMillan - University of Guelph (Canada) [presenting]
Zeny Feng - University of Guelph (Canada)
Lewis Lukens - University of Guelph (Canada)
Abstract: Genomic selection (GS) is a powerful tool to predict the phenotype of an organism based solely on its genotype. GS has quickly become integral to large-scale crop breeding programs, as the ability to fit genome prediction (GP) models using high-density markers can accelerate the rate of genetic gain. A limiting factor in the application of GS to crop breeding is the genotype-by-environment interaction effect. This effect confounds the ability of GP models to identify elite lines in a breeding scheme, slowing genetic gain. The aim is to present a novel framework to elucidate the temporal dimension of the genotype-by-environment interaction effect through the use of weekly climate observations, as well as static soil and landscape factors, along with directly incorporating single-nucleotide polymorphism (SNP) data in a modified temporal fusion transformer (TFT) model developed for the prediction of previously unseen crop varieties to emulate a breeding scheme. It is demonstrated that the TFT model is not only able to significantly outperform linear models commonly used in the field, but it can also provide previously unseen insight into the effects of individual climate and SNP variables across the length of a growing season.
