Back

Covariate-aware genomic prediction of blood metabolite profiles using multi-task neural networks

Guler, M. N.; Alver, M.; Haller, T.; Jay, F.; Pagani, L.; Milani, L.; Yelmen, B.

2026-06-11 genomics
10.64898/2026.06.08.728708 bioRxiv
Show abstract

Circulating metabolites capture clinically relevant physiological variation and contribute to disease aetiology yet are mostly studied as biomarkers rather than prediction targets. Although genome-wide association studies have identified genetic determinants of metabolites, marginal associations do not show how accurately the metabolome can be predicted or whether deep learning approaches can improve prediction by exploiting nonlinearities and dependencies. Here, we developed a multi-task neural network (NN) for predicting metabolomic profiles with a three-stage architecture separating covariate, genotype and joint covariate-genotype 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 driven by nonlinear covariate modelling, with limited and heterogeneous genetic and joint covariate-genotype contributions. With further validation, multi-task NNs could serve as compact, efficient models for predicting metabolite profiles in research and clinical settings.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.