Variational Bayesian Multi-Output Gaussian Process Regression for Metabolic Profiles Prediction With Microbiome Data

Understanding the pivotal role of the human microbiome in health necessitates accurate metabolite prediction, which is crucial for unraveling the intricate interplay between the gut microbiome and human health. This study introduces an innovative approach, Variational Bayesian Multi-Output Gaussian Process Regression (VBMOGPR), to address the challenges posed by the complex, high-dimensional nature of microbiome data. VBMOGPR predicts microbial metabolites, quantifies the model confidence, and incorporates uncertainty estimates. Employing a Bayesian framework with Automatic Relevance Determination (ARD) for feature selection enhances interpretability and performance. Comparative analysis across 14 datasets within a meta-database demonstrated the superiority of VBMOGPR, marking a significant advancement in metabolite prediction and its implications for microbiome impact on human health. In addition, we confirmed that VBMOGPR could tap the potential microbial metabolic association.