Enhancing the potential of phenomic and genomic prediction in winter wheat breeding using high-throughput phenotyping and deep learning

Authors: KAUSHAL, SWAS - North Dakota State University; GILL, HARSIMARDEEP - North Dakota State University; BILLAH, MOHAMMAD - North Dakota State University; KHAN, SHAHID - North Dakota State University; HALDER, JYOTIRMOY - North Dakota State University; Bernardo, Amy; St Amand, Paul; Bai, Guihua; GLOVER, KARL - North Dakota State University; MAIMAITIJIANG, MAITINIYAZI - North Dakota State University; SEHGAL, SUNISH - North Dakota State University

Submitted to: Frontiers in Plant Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/6/2024
Publication Date: 5/30/2024
Citation: Kaushal, S., Gill, H., Billah, M., Khan, S., Halder, J., Bernardo, A.E., St Amand, P.C., Bai, G., Glover, K., Maimaitijiang, M., Sehgal, S.K. 2024. Enhancing the potential of phenomic and genomic prediction in winter wheat breeding using high-throughput phenotyping and deep learning. Frontiers in Plant Science. 15. Article 1410249. https://doi.org/10.3389/fpls.2024.1410249.
DOI: https://doi.org/10.3389/fpls.2024.1410249

Interpretive Summary: Integrating high-throughput phenotyping (HTP) into phenomic and genomic selection (GS) can accelerate breeding for high-yielding wheat. In this study, significant correlations were observed between the HTP-based traits and agronomic traits across growth stages. HTP-based phenomic predictions using a deep neural network (DNN) model showed high accuracies, with the highest GY prediction accuracies at milky ripe stage in the advanced breeding materials. Furthermore, prediction of GY using DNN trained multi-location data from advanced breeding lines improved the accuracy by 32% in preliminary breeding materials when compared to a single location. We evaluated the incorporation of HTP traits into the MT-GS models to predict GY, GPC, and TW. Multi-trait models demonstrated higher prediction accuracy than the single-trait model for GY. Integrating HTP traits into deep-learning-based phenomic or multi-trait genomic selection models will enhance breeding efficiency.

Technical Abstract: Integrating high-throughput phenotyping (HTP) traits into phenomic and genomic selection (GS) can accelerate breeding of high-yielding and climate-resilient wheat cultivars. In this study, we explored the usability of HTP combined with deep learning (DL) for the phenomic or multi-trait (MT) genomic prediction of grain yield (GY), grain protein (GPC), and test weight (TW) in winter wheat. Significant correlations were observed between agronomic traits and HTP traits across different growth stages. Using a deep neural network (DNN) model, HTP-based phenomic predictions showed robust prediction accuracies for GY, GPC, and TW for single locations (0.71, 0.49, and 0.62, respectively) and multiple locations (0.77, 0.73, and 0.63, respectively) using advanced breeding materials. The prediction accuracies for GY varied across growth stages, with Feekes 11 (Milky Ripe) yielding the highest accuracy. Furthermore, forward prediction of GY in preliminary breeding materials using DNN trained on multi-location data from advanced breeding lines improved the accuracy by 32% when compared to a single location. Next, we evaluated the incorporation of HTP traits into the MT-GS models to predict GY, GPC, and TW. MT models using the anthocyanin reflectance index (ARI), green chlorophyll index (GCI), and ratio vegetation index (RVI) as covariates demonstrated higher prediction accuracy (R2 = 0.40, 0.40, and 0.37, respectively) than the single-trait model (R2 = 0.23) for GY. Overall, this study underscores the potential of integrating HTP traits into DL-based phenomic or MT genomic selection models to enhance breeding efficiency.