An AI Fusion Model Based on Molecular Docking and Large Language Models—— Used to Predict the Interaction Between Traditional Chinese Medicine and Disease

Traditional Chinese Medicine (TCM) is a profound and sophisticated medical system. However, the complexity of multi-component interactions and limited computational methods pose significant challenges for optimizing TCM formulations and developing new drugs. This study aims to develop an integrated model that combines molecular docking technology with large language models to achieve high-throughput prediction of interactions between TCM herbs and diseases. We first collected active components from 500 TCM herbs in the TCMSP database and obtained 3D structural information on 100 disease-related targets from the RCSB PDB database. Using OpenBabel to convert SMILES strings into 3D molecular structures, we performed molecular docking calculations with AutoDock Vina. The study defined effective binding as interactions with binding energies ≤-7.0 kcal/mol, yielding 12,408 valid herb-target pairs. On the basis of these data, we trained a Transformer-based neural network model for predicting new TCM–disease interactions. The experimental results demonstrated that the integrated model achieved excellent performance, with an AUC of 0.984 and an AUPR of 0.982 on the test set, significantly outperforming standalone molecular docking or machine learning methods. This proposed integrated model can substantially accelerate modernization research in TCM, providing a powerful tool for elucidating TCM mechanisms and advancing drug development.