Gangloff, H. | bioRxiv Science

bioRxiv2024

Global Genotype by Environment Prediction Competition Reveals That Diverse Modeling Strategies Can Deliver Satisfactory Maize Yield Estimates

Predicting phenotypes from a combination of genetic and environmental factors is a grand challenge of modern biology. Slight improvements in this area have the potential to save lives, improve food and fuel security, permit better care of the planet, and create other positive outcomes. In 2022 and 2023 the first open-to-the-public Genomes to Fields (G2F) initiative Genotype by Environment (GxE) prediction competition was held using a large dataset including genomic variation, phenotype and weather measurements and field management notes, gathered by the project over nine years. The competition attracted registrants from around the world with representation from academic, government, industry, and non-profit institutions as well as unaffiliated. These participants came from diverse disciplines include plant science, animal science, breeding, statistics, computational biology and others. Some participants had no formal genetics or plant-related training, and some were just beginning their graduate education. The teams applied varied methods and strategies, providing a wealth of modeling knowledge based on a common dataset. The winners strategy involved two models combining machine learning and traditional breeding tools: one model emphasized environment using features extracted by Random Forest, Ridge Regression and Least-squares, and one focused on genetics. Other high-performing teams methods included quantitative genetics, classical machine learning/deep learning, mechanistic models, and model ensembles. The dataset factors used, such as genetics; weather; and management data, were also diverse, demonstrating that no single model or strategy is far superior to all others within the context of this competition.

genetics↗

bioRxiv2024

Predicting nonlinear genetic relationships between traits in multi-trait evaluations by using a GBLUP-assisted Deep Learning model

BackgroundGenomic prediction aims to predict the breeding values of multiple complex traits, usually assumed to be normally distributed by the largely used statistical methods, thus imposing linear genetic correlations between traits. While statistical methods are of great value for genomic prediction, these methods do not account for nonlinear genetic relationships between traits. If such relationships exist, although statistical models do perform a fair linear approximation, their prediction accuracy is limited due to the nonlinearity. Deep learning (DL) is a promising methodology for predicting multiple complex traits, in scenarios where nonlinear genetic relationships are present, due to its capacity to capture complex and nonlinear patterns in large data. We proposed a novel hybrid DLGBLUP model which uses the output of the traditional GBLUP, and enhances its PGV by accounting for nonlinear genetic relationships between traits using DL. Using simulated data, we compared the accuracy of the PGV obtained with the proposed hybrid DLGBLUP model, a DL model, and the traditional GBLUP model - the latter being our baseline reference. ResultsWe found that both DL and DLGBLUP models either outperformed GBLUP, or presented equally accurate PGV, with a particular greater accuracy for traits presenting a strongly characterized nonlinear genetic relationship. Overall, DLGBLUP presented the highest prediction accuracy, up to 0.2 points higher than GBLUP, and smallest mean squared error of the PGV for all traits. Additionally, we evolved a base population over seven generations and compared the genetic progress when selecting individuals based on the additive PGV obtained by either DL, DLGBLUP or GBLUP. For all traits with a nonlinear genetic relationship, after the fourth generation, the observed genetic gain when selection was based on the additive PGV from GBLUP was always inferior to the one achieved from either DL or DLGBLUP. ConclusionsThe integration of DL into genomic prediction enables the possibility of modeling nonlinear relationships between traits. Moreover, by identifying these nonlinear genetic relationships, our DL and DLGBLUP models improved prediction accuracy, when compared to GBLUP. The possibility of nonlinear relationships between traits offers a different perspective into multi-trait evaluations and prediction, as well as into the traits evolution over generations, with potential to further improve selection strategies in commercial livestock breeding programs. Moreover, DLGBLUP shows that DL can be used as a complement to statistical methods, by enhancing their performance.

genomics↗