PMPIHGLL: predicting metabolite-protein interactions using dual hypergraph convolutional networks and large language models.

PMPIHGLL achieves over 0.9 AUC and AUPR in predicting metabolite-protein interactions across multiple datasets, outperforming existing models.

• Why it matters: Accurate MPI prediction is crucial for understanding cellular processes and advancing drug discovery, yet current methods lack the ability to capture complex higher-order relationships.
• What they did: The approach integrates large language models for feature extraction, constructs dual hypergraph convolutional networks with different K-nearest-neighbor hypergraphs, and employs attention mechanisms and CNNs for prediction, tested on four datasets with cross-validation.
• The result: This model demonstrates robust performance, including on unseen metabolites and proteins, enabling more reliable identification of MPIs and supporting systems biology and pharmacological research.