Improving Genomic Prediction Using High-Dimensional Secondary Phenotypes

Abstract

In the past decades, genomic prediction has had a large impact on plant breeding. Given the current advances of high-throughput phenotyping and sequencing technologies, it is increasingly common to observe a large number of traits, in addition to the target trait of interest. This raises the important question whether these additional or “secondary” traits can be used to improve genomic prediction for the target trait. With only a small number of secondary traits, this is known to be the case, given sufficiently high heritabilities and genetic correlations. Here we focus on the more challenging situation with a large number of secondary traits, which is increasingly common since the arrival of high-throughput phenotyping. In this case, secondary traits are usually incorporated through additional relatedness matrices. This approach is however infeasible when secondary traits are not measured on the test set, and cannot distinguish between genetic and non-genetic correlations. An alternative direction is to extend the classical selection indices using penalized regression. So far, penalized selection indices have not been applied in a genomic prediction setting, and require plot-level data in order to reliably estimate genetic correlations. Here we aim to overcome these limitations, using two novel approaches. Our first approach relies on a dimension reduction of the secondary traits, using either penalized regression or random forests (LS-BLUP/RF-BLUP). We then compute the bivariate GBLUP with the dimension reduction as secondary trait. For simulated data (with available plot-level data), we also use bivariate GBLUP with the penalized selection index as secondary trait (SI-BLUP). In our second approach (GM-BLUP), we follow existing multi-kernel methods but replace secondary traits by their genomic predictions, with the advantage that genomic prediction is also possible when secondary traits are only measured on the training set. For most of our simulated data, SI-BLUP was most accurate, often closely followed by RF-BLUP or LS-BLUP. In real datasets, involving metabolites in Arabidopsis and transcriptomics in maize, no method could substantially improve over univariate prediction when secondary traits were only available on the training set. LS-BLUP and RF-BLUP were most accurate when secondary traits were available also for the test set.

Integrating and optimizing genomic, weather, and secondary trait data for multiclass classification

“…There is limited literature ( Schrag et al, 2018 ; Akdemir et al, 2020 ; Lopez-Cruz et al, 2020 ; Arouisse et al, 2021 ; Sandhu et al, 2021 ) on combining data types to improve genomic selection. With the advances in modern plant breeding and access to an increasing number of data sources, it is essential to develop statistical approaches that will allow breeders to leverage all available data to improve selection strategies and accelerate breeding programs.…”

“…The SI was used in a G-BLUP model as a covariate or using bivariate methods with SI and main trait as the two responses. Arouisse et al (2021) examined the possibility of other dimensionality reduction methods such as penalized regression and random forests to reduce the dimension of the secondary trait set and used them in bivariate or multivariate settings. Sandhu et al (2021) explored the possibility of including all secondary traits along with the main trait in multivariate GS methods and hence their approach was optimal in the presence of a small number of secondary traits.…”

Recent advances in artificial intelligence, mechanistic models, and speed breeding offer exciting opportunities for precise and accelerated genomics‐assisted breeding

“…This integration of higher‐order gene interactions significantly increased the rice yield prediction from 0.159 (GP alone) to 0.245 (multilayered LASSO); here, GP alone represents the normal GP involving two datasets, viz., genotypic and phenotypic (Hu et al, 2019). Integrating genotypic data with omics features from transcriptomic, metabolomic, and proteomic data using AI‐based models has also shown enhanced phenotypic prediction, leading to improved predictability of yield‐related traits in hybrid rice (Arouisse et al, 2021; Wang et al, 2022). Moreover, DL methods, with their ability to incorporate different features as different layers of a neural network, offer a potential solution for integrating genotypes, environmental factors, and omics data for the genomic prediction of phenotypes (Shook et al, 2021), leading to increased prediction accuracy and appropriate cultivar selection for specific or multiple environments.…”

Knowledge-driven approaches to improve genomic prediction in plants

“…For example, multiple spectral reflectance indices, representing plant physiological and biochemical characteristics, can be calculated using spectral radiometry and used as secondary traits (Sandhu, Mihalyov et al 2021). In addition, secondary traits preferably have higher heritabilities than the primary trait (Velazco, Jordan et al 2019, Bhatta, Gutierrez et al 2020, Arouisse, Theeuwen et al 2021) and share sufficient genetic variation. The level of shared genetic variation, called genetic correlation, indicates the extent to which MT models can improve over ST. For instance, in case of a single secondary trait, the accuracy on a target trait improves when its heritability is lower than the heritability of the secondary trait times the squared genetic correlation (Schulthess, Wang et al 2016, Arouisse, Theeuwen et al 2021.…”

“…Many studies have assessed the potential of MT-GP, using both simulated and real traits, using different approaches. For instance, secondary traits can be modelled as an additional random effect of their phenotypic values in a univariate mixed linear model (Azodi, Pardo et al 2020, Arouisse, Theeuwen et al 2021. Alternatively, a multivariate model can be used to jointly model all traits with a joint distribution accounting for their genetic (co)variance (Bhatta, Gutierrez et al 2020).…”