OptimGS: a dual integrative genomic prediction framework for improving cold stress tolerance in wheat

Cold stress tolerance in wheat is a complex quantitative trait with low heritability, posing significant challenge for conventional breeding programs. Genomic selection offers a powerful framework for accelerating genetic gain; however, its prediction accuracy remains highly dependent on model choice and underlying genetic architecture. In this study, we propose a dual integrative genomic prediction framework designed to enhance prediction accuracy by sequentially integrating information across chromosomes and across models. Using a diverse wheat germplasm panel of 4269 genotypes evaluated for seedling cold tolerance over 2 years, we implemented 14 genomic prediction models spanning Bayesian, best linear unbiased prediction-based, and machine learning approaches. Genome-wide markers were first partitioned chromosome-wise, and predictions were generated independently for each chromosome. These predictions were then optimally combined using genetic algorithm under two bidirectional strategies: chromosome-first-model-second (CFMS) and model-first-chromosome-second (MFCS). Prediction performance was assessed through repeated five-fold cross-validation schemes, with Pearson's correlation coefficient and mean squared error as performance metrics. The CFMS and MFCS strategies consistently outperformed individual models and conventional whole-genome approaches across all datasets. Overall, the proposed framework provides a robust and biologically meaningful strategy for improving genomic prediction of complex quantitative traits and holds potential for accelerating crop improvement programs. The source code of the developed framework is available at https://github.com/PrabinaMeher/OptimGS.git.