Application Of Machine Learning In Plant Breeding

Genomic selection (GS) is a new method of selective breeding using high-density markers covering the whole genome, in short, marker-assisted selection in the whole genome. As a new generation of breeding technology, genome-wide selection can shorten the breeding generation interval, accelerate the breeding process, save costs, and promote the development of modern breeding in the direction of precision and high efficiency by constructing prediction models and predicting and selecting early individuals based on genome-wide estimated breeding values. Statistical models, as the core of genome-wide selection, greatly affect the accuracy and efficiency of genome-wide prediction. Traditional prediction methods are based on linear regression models, which are difficult to capture the complex relationship between genotypes and phenotypes. Compared with traditional models, non-linear models have the ability to analyse complex non-additive effects, and Machine Learning (ML) and Deep Learning (DL) algorithms provide new opportunities to solve the challenges of big data analysis and high-performance parallel computing.

In order to respond to the learning needs of the majority of students in this direction, Beijing Jinzhi Research will hold the third Whole Genome Selection Learning Exchange on 28~30 June. At that time, we will invite researcher Li Huihui, the chief of Big Data Intelligent Design and Breeding Innovation Team of Institute of Crop Science, Chinese Academy of Agricultural Sciences (CAAS), and doctoral supervisor, to give us a systematic lecture on genome-wide selection based on machine learning and deep learning algorithms. Ms Li has long been engaged in bioinformatics, statistical genomics and quantitative genetics related researches, such as research and development of breeding prediction algorithms based on high-throughput genomics data, and model construction; she has published first/first papers in Molecular Plant, Nature Plants, Plant Biotechnology Journal, Trends in Plant Science and other journals. He has published 45 articles in Molecular Plant, Nature Plants, Plant Biotechnology Journal, Trends in Plant Science and other journals as the first/corresponding author, and his research results have won the Top Ten Outstanding Agricultural Science and Technology Achievements of Hainan Province in the year of 2023; he has been awarded the Outstanding Youth Science Fund of the National Foundation of China, and the Youth Science and Technology Award of the Chinese Agricultural Society; he has been selected as a leading talent of the Chinese Academy of Agricultural Sciences (CASA). He is also the Secretary General of the Intelligent Agriculture Committee of the Crop Society of China.

RF, XGBoost, GBDT, LightGBM), genome-wide selection models based on deep learning algorithms (DNNGP) to implement genome prediction; model optimisation and implementation of target group prediction; and in particular, compared to the previous two issues, the addition of a new deep learning model for prediction of functional genes based on high-throughput genome data as well as a big language model. On this basis, Ms Li's team will do detailed prediction demonstration and cross-validation result evaluation and visualisation for each model, and take you to build deep learning network using KNIME platform, which will eventually allow us to carry out genome-wide prediction based on deep learning algorithms using genomics big data. The class will last for three days, aiming at solving the pain points and difficulties we encountered in the process of practice, and I believe it will present an audio-visual feast of genome-wide selection for everyone.