Multi-Trait Machine and Deep Learning Models for Genomic Selection using Spectral Information in a Wheat Breeding Program

Abstract

Prediction of breeding values and phenotypes is central to plant breeding and has been revolutionized by the adoption of genomic selection (GS). Use of machine and deep learning algorithms applied to complex traits in plants can improve prediction accuracies in the context of GS. Spectral reflectance indices further provide information about various physiological parameters previously undetectable in plants. This research explores the potential of multi-trait (MT) machine and deep learning models for predicting grain yield and grain protein content in wheat using spectral information in GS models. This study compares the performance of four machine and deep learning -based uni-trait (UT) and MT models with traditional GBLUP and Bayesian models. The dataset consisted of 650 recombinant inbred lines from a spring wheat breeding program, grown for three years (2014-2016), and spectral data were collected at heading and grain filling stages. MT-GS models performed 0-28.5% and -0.04-15% superior to the UT-GS models for predicting grain yield and grain protein content. Random forest and multilayer perceptron were the best performing machine and deep learning models to predict both traits. These two models performed similarly under UT and MT-GS models. Four explored Bayesian models gave similar accuracies, which were less than machine and deep learning-based models, and required increased computational time. Green normalized difference vegetation index best predicted grain protein content in seven out of the nine MT-GS models. Overall, this study concluded that machine and deep learning-based MT-GS models increased prediction accuracy and should be employed in large-scale breeding programs.

Genomic Selection for End-Use Quality and Processing Traits in Soft White Winter Wheat Breeding Program with Machine and Deep Learning Models

“…Due to the unbalanced nature of the dataset, adjusted means were calculated using residuals obtained using the lme4 R package for within environment analysis. The model equation is represented as Y ij = Block i + Check j + e ij (1) where Y ij is the raw phenotype; Check j is the effect of replicated check cultivar; Block i corresponds to the fixed block effect; and e ij is the residuals [30,31]. Block was considered fixed, as we want to remove that component of variation before exploring the genetic variation.…”

“…Regularization, dropout, and early stopping were applied to control overfitting. Furthermore, information about hyperparameter optimization and deep learning models is referred to in [21,31].…”

“…CNN is a special case of a deep learning model that accounts for the specific pattern present between the input features. Information about the CNN model, its implementation, and hyperparameter optimization are referred to in previous publications [21,31]. A combination of input, convolutional, pooling, dense, flatten, dropout, and output layers were applied for the prediction.…”

Genome-Wide Association Studies and Genomic Selection for Grain Protein Content Stability in a Nested Association Mapping Population of Spring Wheat

“…Models trained on 2014 GPC data were used for predictions in 2015 and 2016. Similarly, the 2015 GPC training model was used for 2016 predictions (Sandhu et al 2021c).…”

Phenomic data-facilitated rust and senescence prediction in maize using machine learning algorithms

“…The use of VIs and canopy temperature as predictors were shown to increase model accuracy of genomic and hybrid model accuracies for wheat grain yield 65 . Sandhu et al 66 suggested using spectral information as a secondary trait in genomic prediction provided better prediction accuracy for grain protein content. Galán et al 67 argue that genomic models incorporate genetic relationships between untested candidates and those with known genotypic and phenotypic information, and their study revealed hyperspectral reflectance-derived (HBLUP) relationship matrices (i.e., HTP data) were less prone to genetic relatedness and trait heritability, whereas more highly heritable traits were better predicted by genomic (GBLUP) relationship matrices.…”