Covariate-aware genomic prediction of blood metabolite profiles using multi-task neural networks
- 1Institute of Genomics, University of Tartu
Abstract
Predictive models of quantitative traits can combine genetic and covariate information, but overall predictive performance alone does not reveal whether differences between models arise from covariate, genetic or joint covariate-genetic effects. Circulating metabolites provide a high-dimensional set of clinically relevant quantitative traits in which these effects can be examined systematically. Although their genetic determinants are well characterised through genome-wide association studies, marginal associations do not establish how accurately metabolomic profiles can be predicted or whether nonlinear models can improve prediction by capturing complex and potentially interactive structure. Here, we developed a multi-task neural network (NN) for simultaneously predicting metabolomic profiles with a three-stage architecture separating covariate, genetic and joint contributions. In comparative analyses, the multi-task NN demonstrated the strongest mean performance across metabolites (R2=0.219), followed by the single-task NN (R2=0.211), elastic net (R2=0.207), and an activation-free multi-task model (R2=0.191). Decomposition analyses indicated that gains were mainly driven by nonlinear covariate modelling, consistent with improved prediction after incorporating nonlinear age effects into a linear model. Together, these analyses provide a framework for identifying sources of predictive differences between different models, which may be applicable to other collections of quantitative traits.
Related
- Integrating longitudinal metabolite profiles improves phenotype prediction in pigs in a trait- and timepoint-dependent manner
- Exploratory analysis of genomic prediction profiles in Arequipa fighting cattle using commercial SNP panels
- Substrate-driven shifts in microbial communities reshape metabolite profiles during soybean residue fermentation
- Metabolomics Reveals Differences in Muscle Metabolite Profiles Between Daweishan Mini Chickens and Tegel Broilers at Marketable Age.
- UHPLC-TIMS-qTOF-MS/MS expands metabolite annotation and identifies distinct cultivar-specific metabolite profiles in American elderberry flowers
BibTeX
@misc{guler2026covariateaware,
title = {{Covariate-aware genomic prediction of blood metabolite profiles using multi-task neural networks}},
author = {M. N. Guler and M. Alver and T. Haller and F. Jay and L. Pagani and L. Milani and B. Yelmen},
howpublished = {bioRxiv},
year = {2026},
month = jun,
doi = {10.64898/2026.06.08.728708},
url = {https://www.biorxiv.org/content/10.64898/2026.06.08.728708v2},
note = {CC BY},
}
RIS
TY - UNPB
TI - Covariate-aware genomic prediction of blood metabolite profiles using multi-task neural networks
AU - M. N. Guler
AU - M. Alver
AU - T. Haller
AU - F. Jay
AU - L. Pagani
AU - L. Milani
AU - B. Yelmen
PY - 2026
DA - 2026/06/08
PB - bioRxiv
DO - 10.64898/2026.06.08.728708
UR - https://www.biorxiv.org/content/10.64898/2026.06.08.728708v2
AB - Predictive models of quantitative traits can combine genetic and covariate information, but overall predictive performance alone does not reveal whether differences between models arise from covariate, genetic or joint covariate-genetic effects. Circulating metabolites provide a high-dimensional set of clinically relevant quantitative traits in which these effects can be examined systematically. Although their genetic determinants are well characterised through genome-wide association studies, marginal associations do not establish how accurately metabolomic profiles can be predicted or whether nonlinear models can improve prediction by capturing complex and potentially interactive structure. Here, we developed a multi-task neural network (NN) for simultaneously predicting metabolomic profiles with a three-stage architecture separating covariate, genetic and joint contributions. In comparative analyses, the multi-task NN demonstrated the strongest mean performance across metabolites (R2=0.219), followed by the single-task NN (R2=0.211), elastic net (R2=0.207), and an activation-free multi-task model (R2=0.191). Decomposition analyses indicated that gains were mainly driven by nonlinear covariate modelling, consistent with improved prediction after incorporating nonlinear age effects into a linear model. Together, these analyses provide a framework for identifying sources of predictive differences between different models, which may be applicable to other collections of quantitative traits.
ER -