A1187
Title: Bayesian optimization of genotype and environment interaction
Authors: Hiroyoshi Iwata - The University of Tokyo (Japan) [presenting]
Tai-Shen Chen - The University of Tokyo (Japan)
Chikashi Sato - Ifurinrin (Japan)
Masanori Yamasaki - Niigata University (Japan)
Chyon Hae Kim - Iwate University (Japan)
Akira Abe - Iwate Biotechnology Research Center (Japan)
Hiroyuki Shimono - Iwate University (Japan)
Abstract: Stable food production requires the development of new varieties that can adapt to changes in the environment. Genomic selection can accelerate the development of such varieties. Note that since adaptation to future environments is a statistical extrapolation, the interaction between genotype and environment must be optimized, taking into account the uncertainty associated with the extrapolation. A Bayesian optimization (BO) method for genotype-environment interactions is proposed, using the analysis of historical Japanese rice breeding data as an example. Phenotypic variation caused by the combination of genotype and environment was modelled using a Gaussian process (GP), and simulations were performed using BO to search for the optimal combination of genotype and environment in the region to be extrapolated. The results showed that BO was able to detect the optimal combination more efficiently than point prediction in regions with high extrapolation. In addition, the optimal BO criterion was found to differ depending on the region of interest. In summary, it is found that the approach using modelling by GP and BO was effective in searching for the optimal combination of genotype and environment.