Machine Learning Solutions for Integrating Partially Overlapped Genetic Datasets and Modelling Host-Endophyte Effects in Estimating Ryegrass (Lolium) Dry Matter Yield

Jiashuai Zhu ^1,2* M Michelle Malmberg ² Maiko Shinozuka ² Renata M Retegan ² Noel Cogan ^2,3 Joe L Jacobs ^1,2 Khageswor Giri ² Kevin F Smith ^1,2

• ¹ The University of Melbourne, Parkville, Australia
• ² Department of Energy, Environment and Climate Action, Victoria State Government, Melbourne, Victoria, Australia
• ³ AgriBio, La Trobe University, Melbourne, Victoria, Australia

Plant genetic evaluation often faces challenges due to complex genetic structures. Ryegrass (Lolium), a valuable species for pasture-based agriculture, is one such plant species that presents heterogeneous genetic diversities among base breeding populations. Partially overlapped datasets arising from incompatible studies and commercial restrictions further impede the integration of outcomes across studies and challenge the evaluation of valuable agricultural traits, such as dry matter yield (DMY). As these, 1) we implemented a population genotyping approach to capture the genetic diversity in ryegrass base cultivars; 2) we introduced a machine learning based strategy to integrate genetic distance matrices (GDMs) from incompatible genotyping approaches, including alignments using multidimensional scaling (MDS) and Procrustes transformation, as well as a novel evaluation strategy (BESMI) for the imputation of structural missing data. Endophytes complicate genetic evaluation by introducing additional variation in phenotypic expression. 3) We modelled the impacts of nine commercial endophytes on ryegrass DMY, enabling a more balanced estimation for untested cultivar-endophyte combinations. 4) The phylogenetic analysis offered a pseudo-pedigree relationship of the 113 ryegrass populations and revealed its associations with DMY variations. Overall, this research provides practical insights to integrate partially overlapped GDMs with structural missing data patterns and facilitates the identification of high-performing ryegrass clades. The methodological advancements, including population sequencing, MDS alignment via Procrustes transformation, and BESMI, have applications beyond ryegrass.