Identification of Genes Accurately Predicting Fiber Length for AI-Powered Gene-Based Breeding Across Populations and Environments in Cotton.

Abstract

AI-powered gene-based breeding (AI-GBB) has the potential to be a next-generation breeding technology. However, it is unclear whether agronomic genes identified with one population can be reliably used for AI-GBB across diverse populations, environments, and eco-agriculture systems. We evaluated 474 cotton fiber length (GFL) genes using two diverse populations grown across 11 environments in China and the United States. Four hundred and seventy-five functional SNPs significantly impacting fiber length were identified from 143 of the genes and tested for AI-GBB. Either the 475 functional SNPs or their 143 SNPs with one SNP per gene predicted fiber lengths across environments using either the total number of their favorable alleles or themselves as DNA markers at accuracies of 0.75-0.89, outperforming genomic selection by several times. The prediction model trained in one environment accurately and consistently predicted fiber length in all other environments. We found that the GFL genes enriched in proteolysis, protein binding, and UDP-sugar pyro-phosphorylases play important roles in their predictability. Our results demonstrate that agronomic genes identified from one population can be robust predictors of quantitative traits across diverse populations and environments for AI-GBB. The functional SNPs provide the most desirable tools for enhanced cotton breeding with all molecular breeding methods developed to date, and the utility knowledge of the genes is necessary for the development and deployment of AI-GBB in all crops.