An evaluation of machine-learning for predicting phenotype: studies in yeast, rice, and wheat

Abstract

In phenotype prediction the physical characteristics of an organism are predicted from knowledge of its genotype and environment. Such studies, often called genome-wide association studies, are of the highest societal importance, as they are of central importance to medicine, crop-breeding, etc. We investigated three phenotype prediction problems: one simple and clean (yeast), and the other two complex and real-world (rice and wheat). We compared standard machine learning methods; elastic net, ridge regression, lasso regression, random forest, gradient boosting machines (GBM), and support vector machines (SVM), with two stateof-the-art classical statistical genetics methods; genomic BLUP and a two-step sequential method based on linear regression. Additionally, using the clean yeast data, we investigated how performance varied with the complexity of the biological mechanism, the amount of observational noise, the number of examples, the amount of missing data, and the use of different data representations. We found that for almost all the phenotypes considered, standard machine learning methods outperformed the methods from classical statistical genetics. On the yeast problem, the most successful method was GBM, followed by lasso regression, and the two statistical genetics methods; with greater mechanistic complexity GBM was best, while in simpler cases lasso was superior. In the wheat and rice studies the best two methods were SVM and BLUP. The most robust method in the presence of noise, missing data, etc. was random forests. The classical statistical genetics method of genomic BLUP was found to perform well on problems where there was population structure. This suggests that standard machine learning methods need to be refined to include population structure information when this is present. We conclude that the application of machine learning methods to phenotype prediction problems holds great promise, but that determining which methods is likely to perform well on any given problem is elusive and non-trivial.

Potential of Genome-Wide Association Studies and Genomic Selection to Improve Productivity and Quality of Commercial Timber Species in Tropical Rainforest, a Case Study of Shorea platyclados

“…GS is fundamentally different from GWAS, as it involves use a full-genome information, regardless of its significance, in relation to a specific trait, rather than a few markers as in GWAS. This genotypic information, collected from training and validation population, is used in conjunction with corresponding phenotypic data, collected from training population, to develop a predictive model [12,14]. In forest tree breeding programs, GWAS and GS could substantially reduce the length of breeding cycles and increase genetic gain per unit time through early selection of superior genotypes during the juvenile phase.…”

“…Since these statistical methods cannot explicitly account for interactions among single nucleotide polymorphisms (SNPs), application of Machine Learning in GS studies has been proposed. Machine Learning is being increasingly applied in GS studies because it does not require any assumptions about the underlying traits, it is easy to use, and it can both capture complex non-linear relationships and efficiently increase prediction accuracy [14]. Popular Machine Learning methods include Random Forest (RF), Extreme Gradient Boosting (XgBoost) and Bayesian Additive Regression Tree (BART) modelling.…”

Multi-tissue transcriptome-wide association studies

“…Previously proposed approaches for learning the imputation rules are based on regularized linear models [11][12][13][14], polygenic risk scores [11] and using the top SNP to predict expression levels [12]. However, the machine learning literature has shown that alternative approaches such as random forests (RF), which allow naturally for non-linear and non-additive effects, can produce more accurate predictions in model organisms [15,16]. We set out to explore whether using RF could also lead to better gene expression predictions in humans and, if so, whether that could be translated into a more powerful TWAS.…”

“…We also sought to take advantage of the fact that expression levels of a given gene in different cell types can be correlated by considering expression values across multiple cell types simultaneously in a multi-task framework. This has been shown to improve multi-trait predictions in yeast [16] and in applications to real and simulated data in marker-assisted selection for several related traits [17][18][19] or populations [20]. Multi-trait approaches have also been used to analyse eQTL datasets [21,22].…”

Predicting rice phenotypes with meta and multi-target learning

“…Machine learning algorithms are increasingly being adapted for the prediction of plant phenotypes (Grinberg et al 2016(Grinberg et al , 2019. This task is most commonly regression based as most agronomic phenotypes are quantitative.…”