Genomic prediction of plant traits by popular machine learning methods

A rapid growth of the available body of genomic data has made it possible to obtain extensive results in genomic prediction and identification of associations of SNPs with phenotypic traits. In many cases, to identify new relationships between phenotypes and genotypes, it is preferable to use machine learning, deep learning and artificial intelligence, especially explainable artificial intelligence, capable of recognizing complex patterns. 80 sources were manually selected; while there were no restrictions on the release date, the main attention was paid to the originality of the proposed approach for use in genomic prediction. The article considers models for genomic prediction, convolutional neural networks, explainable artificial intelligence and large language models. Attention is paid to Data Augmentation, Transfer Learning, Dimensionality Reduction methods and hybrid methods. Research in the field of model-specific and model-independent methods for interpretation of model solutions is represented by three main categories: sensing, perturbation, and surrogate model. The considered examples reflect the main modern trends in this area of research. The growing role of large language models, including those based on transformers, for genetic code processing, as well as the development of data augmentation methods, are noted. Among hybrid approaches, the prospect of combining machine learning models and models of plant development based on biophysical and biochemical processes is emphasized. Since the methods of machine learning and artificial intelligence are the focus of attention of both specialists in various applied fields and fundamental scientists, and also cause public resonance, the number of works devoted to these topics is growing explosively.