Digital Breeding

This research focuses on efficient genomic prediction, which aims to predict phenotypic traits from high-dimensional genomic data. Although Transformer-based models can capture complex relationships among genetic markers, directly processing complete genomic sequences often requires substantial computational resources and may increase the risk of overfitting when the number of samples is limited. 

To address these challenges, we propose SAMoE, a two-stage framework that constructs multiple genomic views through random position selection and combines their predictions using an adaptive mixture-of-experts module. The framework is designed to improve prediction performance while reducing model size, training cost, and inference latency.