Prediction of metabolite-protein intreactions based on integration of machine learning and constraint-based modeling

Abstract Motivation Metabolite-protein interactions play an important role in regulating protein functions and metabolism. Yet, predictions of metabolite-protein using genome-scale metabolic networks are lacking. Here we fill this gap by presenting a computational framework, termed SARTRE, that employs features corresponding to shadow prices determined the context flux variability analysis predict supervised machine learning. Results By gold standards for interactomes well-curated models Escherichia coli Saccharomyces cerevisiae, found implementation SARTRE with random forest classifiers accurately predicts interactions, supported average area under receiver operating curve 0.86 0.85, respectively. Ranking based on their importance classification demonstrated key predicting interactions. The quality is further excellent agreement organism-specific unseen shared between two model organisms. Further, from highly competitive against those obtained recent deep learning approach relying variety metabolite features. Together, these findings show extracted constraint-based analyses pave way understanding functional roles proteins small molecules. Availability https://github.com/fayazsoleymani/SARTRE Supplementary information data available at Bioinformatics online.