Multimodal deep learning methods enhance genomic prediction of wheat breeding

Abstract

While several statistical machine learning methods have been developed and studied for assessing the genomic prediction (GP) accuracy of unobserved phenotypes in plant breeding research, few methods have linked genomics and phenomics (imaging). Deep learning (DL) neural networks have been developed to increase the GP accuracy of unobserved phenotypes while simultaneously accounting for the complexity of genotype × environment interaction (GE); however, unlike conventional GP models, DL has not been investigated for when genomics is linked with phenomics. In this study we used two wheat data sets (DS1 and DS2) to compare a novel DL method with conventional GP models. Models fitted for DS1 were GBLUP, gradient boosting machine (GBM), support vector regression (SVR) and the DL method. Results indicated that for one year, DL provided better GP accuracy than results obtained by the other models. However, GP accuracy obtained for other years indicated that the GBLUP model was slightly superior to the DL. DS2 is comprised only of genomic data from wheat lines tested for three years, two environments (drought and irrigated) and two to four traits. DS2 results showed that when predicting the irrigated environment with the drought environment, DL had higher accuracy than the GBLUP model in all analyzed traits and years. When predicting drought environment with information on the irrigated environment, the DL model and GBLUP model had similar accuracy. The DL method used in this study is novel and presents a strong degree of generalization as several modules can potentially be incorporated and concatenated to produce an output for a multi-trait data structure.

Deep learning methods improve genomic prediction of wheat breeding

“…Yield was measured in all environments, while Germination, Heading, Height and Maturity were determined in three out of four (B5IR, B2IR, and BDRT). Recently this data set was employed by Montesinos-López et al. (2023) for assessing the benefit of applying sparse phenotype field trials for genomic prediction at early testing generation of the population improvement (occurring at F 4 or F 5 )>.…”

“…(2023) . In wheat genomic prediction, multi-modal deep learning models have been explored and applied as a promising approach ( Kick et al., 2023 ; Montesinos-López et al., 2023 ). These studies have demonstrated the potential of multi-modal deep learning in enhancing the accuracy of genomic prediction for wheat traits.…”

“…Based on the previous considerations on how DL can be employed for genomic prediction in this study we follow a similar network structure as the previous study of Montesinos-López et al. (2023) , up to the output layer.…”

Ensemble of BLUP, Machine Learning, and Deep Learning Models Predict Maize Yield Better Than Each Model Alone

“…Combining physiological and statistical models into a single model has also been explored but not yet widely implemented (Messina et al 2018;Diepenbrock et al 2021;Shahhosseini et al 2021a). Among models with substantial environmental, as well as genetic, components, Best Linear Unbiased Predictor models (BLUPs) and deep neural networks (DNNs) have both been shown to perform well under some scenarios (Washburn et al 2021) with a potential tradeoff in average model performance favoring BLUPs and model consistency across replicates in some cases favoring DNNs (Kick et al 2023) in addition to the ability to more directly incorporate multimodal data (Montesinos-López et al 2023). For most breeding applications however, BLUP models with limited or no environmental data are still considered the gold standard with their relative simplicity, when compared to machine learning, deep learning, and other approaches being a major benefit (Ma et al 2018;Montesinos-López et al 2018;Abdollahi-Arpanahi et al 2020;Nazzicari and Biscarini 2022;Gianola et al 2022).…”

“…The combination of genomic and environmental effects (both those within a farmer's control, i.e., management, and those external to it) are necessary because gene by environmental effects can in some cases exceed purely genetic effects . Furthermore, inclusion of these interactions can improve the predictive accuracy for new environments or cultivars (Li et al 2021;Jarquin et al 2021;Montesinos-López et al 2023).…”