A comparison of classical and machine learning-based phenotype prediction methods on simulated data and three plant species

Abstract

Genomic selection is an integral tool for breeders to accurately select plants directly from genotype data leading to faster and more resource-efficient breeding programs. Several prediction methods have been established in the last few years. These range from classical linear mixed models to complex non-linear machine learning approaches, such as Support Vector Regression, and modern deep learning-based architectures. Many of these methods have been extensively evaluated on different crop species with varying outcomes. In this work, our aim is to systematically compare 12 different phenotype prediction models, including basic genomic selection methods to more advanced deep learning-based techniques. More importantly, we assess the performance of these models on simulated phenotype data as well as on real-world data from Arabidopsis thaliana and two breeding datasets from soy and corn. The synthetic phenotypic data allow us to analyze all prediction models and especially the selected markers under controlled and predefined settings. We show that Bayes B and linear regression models with sparsity constraints perform best under different simulation settings with respect to explained variance. Further, we can confirm results from other studies that there is no superiority of more complex neural network-based architectures for phenotype prediction compared to well-established methods. However, on real-world data, for which several prediction models yield comparable results with slight advantages for Elastic Net, this picture is less clear, suggesting that there is a lot of room for future research.

GPFN: Prior-Data Fitted Networks for Genomic Prediction

“…For all results, we compare to GBLUP exclusively, because we consider it to be at least as good as state-of-the-art. It is not our intention to compare multiple classes of models, as this has already been demonstrated extensively in the literature (Azodi et al, 2019; Abdollahi-Arpanahi et al, 2020; Zingaretti et al, 2020; Ubbens et al, 2021; John et al, 2022; Ray et al, 2023). For each experiment, we split the data into 80% training and 20% testing.…”

“…Despite continual progress in genotyping, the prediction of Genomic Estimated Breeding Values (GEBVs) is still most often performed using classical techniques such as BayesB (Meuwissen et al, 2001), and GBLUP (VanRaden, 2008). Much recent work has focused on the discovery of new methods based on deep learning (Ma et al, 2018; Wang et al, 2023; Gao et al, 2023) – however, classical methods remain the most popular because they empirically perform at least as well as newer, more modern methods while being simpler, faster, and requiring less tuning (Azodi et al, 2019; Abdollahi-Arpanahi et al, 2020; Zingaretti et al, 2020; Ubbens et al, 2021; John et al, 2022; Ray et al, 2023).…”

HeliantHOME, a public and centralized database of phenotypic sunflower data

“…The data might also be used for the comparison or development of new phenotype prediction methods. The imaging data can be analyzed using custom Python scripts and might also serve the computer science community to develop novel machine learning and computer vision methods for automatic phenotyping 36,37,38 .…”

easyPheno: An easy-to-use and easy-to-extendPythonframework for phenotype prediction using Bayesian optimization

“…Predicting complex traits from genotypic information remains a major challenge in modern biology, whether to reduce costs and accelerate the breeding process in plants and animals or to assess the risk of diseases in humans. So far, existing studies on phenotype prediction in plants ( John et al , 2022 ), animals ( Abdollahi-Arpanahi et al , 2020 ) and humans ( Bellot et al , 2018 ) fail to determine an overall pre-dominant prediction method. Their results show that the prediction performance is highly dependent on the species-trait combination, thus requiring the re-evaluation of various prediction models.…”