The 11th International Conference on Agricultural and Biological Sciences (ABS 2025)

Abstract: Genome-wide association study (GWAS) is a key gene mining method and has played a significant role in the genetic dissection of complex traits in animals, plants, and humans. However, it still has shortcomings in both theory and application. The genome-wide scanning framework of existing methods does not take advantage of the AI era. Methods based on allele substitution effect (α) have low power to detect dominant, overdominant, rare, and small-α QTNs. There are almost no feasible methods available to detect QTN-by-environment (QEI) and QTN-by-QTN (QQI) interactions. These result in some important loci and trait heritability being missed. To address these issues, we have established a compressed variance component mixed model and a 3VmrMLM method. In this model, all possible effects are considered and all possible polygenic genetic backgrounds are controlled. The model is then used for genome-wide scanning to identify potential QTNs, QEIs, and QQIs. Some of these are then identified as significant QTNs, QEIs, and QQIs using machine learning. This is our two-stage method with a genome-wide scanning plus machine learning framework. This simple model can be used to detect QTNs, QEIs, and QQIs uniformly. Genetic dissection and breeding by design for polygenic traits remain substantial challenges, and its calculation speed still needs to be improved, particularly in the era of big data and AI. Thus, we have integrated advanced statistical and computer technologies with 3VmrMLM to propose Fast3VmrMLM for identifying abundant and key genes for polygenic traits. The advantages of this approach are as follows. In the 18K rice dataset containing ~3 million SNPs, more dominant, small-effect and rare QTNs were identified; 211 known genes and 384 candidate genes for 14 traits were detected; and each trait took 3.30 h to analyze. Furthermore, the SNP marker expanded to include bin/gene haplotypes, lncRNA types, and structural variations, making mQTL detection possible. In the maize NC II breeding population containing over 30 million markers, 26 known genes and 24 candidate genes were found to be associated with seven yield-related traits. These QTNs were then used to predict excellent hybrid combinations. More importantly, a genetic network for rice yield-related traits was constructed using all the known and candidate genes. The key genes in this network were identified and a breeding strategy utilizing these key genes was proposed. The software packages IIIVmrMLM and Fast3VmrMLM can be downloaded for free from https://github.com/YuanmingZhang65. Additionally, the new methods for QEI and QQI detection in big data were introduced briefly.

Biography: Yuan-Ming Zhang is a professor of statistical genomics at Huazhong Agricultural University (HZAU) in China. His major is quantitative genetics. He obtained his Bachelor's degree from Southwest Agricultural University in 1986, followed by his Master’s degree and PhD from Nanjing Agricultural University (NAU) in 1992 and 2001 respectively. He worked on the Rongchang campus of Southwest University from 1986 to 1999 before moving to NAU, and then to HZAU in 2014. He was a postdoctoral fellow at the University of California, Riverside, from 2003 to 2005, and became an associate professor in 1995, progressing to full professor in 2002. In 2022, he was selected as one of the world's top 2% of scientists. He is an editorial board member of Heredity, BMC Genomics, Front Plant Sci (Guest), and Acta Agronomica Sinica. He is also a council member of the Chinese Society of Agri-Biotechnology, and a full member of Sigma Xi, the Scientific Research Honor Society, and the Genetics Society UK. His team has developed several software packages for identifying genes associated with complex traits. These include mrMLM, IIIVmrMLM and Fast3VmrMLM for GWAS, GCIM for QTL mapping, dQTG-seq for bulked segregant analysis, and SEA for mixed inheritance analysis. They can be downloaded for free from https://github.com/YuanmingZhang65 and https://cran.r-project.org/web/packages/. He has published over 120 articles in journals such as Mol Plant, Plant Cell, Brief Bioinform and Heredity.