Genomic prediction in plants: opportunities for ensemble machine learning based approaches

Abstract

Background: Many studies have demonstrated the utility of machine learning (ML) methods for genomic prediction (GP) of various plant traits, but a clear rationale for choosing ML over conventionally used, often simpler parametric methods, is still lacking. Predictive performance of GP models might depend on a plethora of factors including sample size, number of markers, population structure and genetic architecture. Methods: Here, we investigate which problem and dataset characteristics are related to good performance of ML methods for genomic prediction. We compare the predictive performance of two frequently used ensemble ML methods (Random Forest and Extreme Gradient Boosting) with parametric methods including genomic best linear unbiased prediction (GBLUP), reproducing kernel Hilbert space regression (RKHS), BayesA and BayesB. To explore problem characteristics, we use simulated and real plant traits under different genetic complexity levels determined by the number of Quantitative Trait Loci (QTLs), heritability (h2 and h2e), population structure and linkage disequilibrium between causal nucleotides and other SNPs. Results: Decision tree based ensemble ML methods are a better choice for nonlinear phenotypes and are comparable to Bayesian methods for linear phenotypes in the case of large effect Quantitative Trait Nucleotides (QTNs). Furthermore, we find that ML methods are susceptible to confounding due to population structure but less sensitive to low linkage disequilibrium than linear parametric methods. Conclusions: Overall, this provides insights into the role of ML in GP as well as guidelines for practitioners.

Genomic basis of seed colour in quinoa inferred from variant patterns using extreme gradient boosting

“…Collinear features were detected using pearsonr from the python library scipy (Virtanen et al ., 2020 ). To test for possible effects of population structure we integrated the top 10 principal components as additional features to the input data set (Farooq et al ., 2022 ; Zhao et al ., 2012 ).…”

Knowledge-driven approaches to improve genomic prediction in plants

“…The prediction accuracy of PRIORNET (Figure 4.2A) on test data is significantly higher (p Wilcoxon < 0.05) than that of the benchmark MLP and all other methods. The accuracy of the parametric methods (Lasso, Elastic Net, BayesA and BayesB) is similar and approaches the narrow-sense heritability (ℎ � = 0.4) as they can only capture the additive effects of SNPs (see Methods) (Farooq, van Dijk et al 2022). The accuracy of the Random Forest is higher, presumably because it can capture interaction effects.…”

“…The GP model is central to GS-based breeding programs and its prediction accuracy determines overall selection accuracy. The accuracy of GP depends on combined effects of population properties, genetic complexity factors and the modelling framework (Farooq, van Dijk et al 2022). Different strategies, such as incorporating population and trait properties as part of the modelling framework, can be employed to improve GP predictive ability.…”