Machine Learning-Driven Prediction of Candidate Umami Compounds: A Structure-Based Interpretation with Application to Pleurotus ostreatus

The umami taste, one of the five basic taste modalities, is primarily mediated by the heterodimeric receptor TAS1R1/TAS1R3, which governs ligand recognition at the molecular level. However, despite this well-established biological framework, the identification of umami-active compounds from natural food sources remains a challenging task due to the chemical complexity of food metabolomes and the limited availability of predictive models targeting this receptor. In the present study, a multi-stage computational pipeline combining machine learning and structure-based analysis was developed and applied for the systematic screening of the Pleurotus ostreatus chemical profile. A curated dataset of 259 compounds with experimentally determined TAS1R1/TAS1R3 activity was used to develop a machine learning (ML) predictive model, with a Linear Discriminant Analysis (LDA) classifier demonstrating the best predictive performance. The applicability of the model was further evaluated using the chemical profile of Pleurotus ostreatus, a mushroom widely associated with pronounced umami characteristics. Selected high-probability compounds were subsequently investigated using molecular docking, enabling the assessment of their binding behavior within a human TAS1R1/TAS1R3 homology model. Amino acid-like compounds exhibited conserved interaction patterns within the predicted orthosteric region, whereas structurally distinct metabolites adopted alternative binding orientations characterized by hydrogen-bonding interactions. The integration of ligand-based prediction with structure-based analysis provides a comprehensive framework for the identification and mechanistic interpretation of umami-related compounds, highlighting the potential of Pleurotus ostreatus as a source of structurally diverse candidate umami-related metabolites.