Investigation of supramolecular self-assembly in traditional Chinese medicine by computer simulations

Traditional Chinese medicine (TCM) supramolecular assemblies are commonly found in TCM decoctions. As a form of the material basis for pharmacological efficacy, TCM supramolecular assemblies not only enhance therapeutic effects and reduce toxicity but also represent an important manifestation of the "integration of medicine and adjuvant" characteristic of TCM components. However, current research rarely focuses on the structural features of TCM components capable of forming supramolecular assemblies, and there is a lack of guiding principles for screening self-assembling components. Evaluating the self-assembly ability of various components solely through classical experiments is time-consuming and labor-intensive. Therefore, in this study, we employed a computational simulation method (N-GAFF-CM) to assess the self-assembly capabilities of 55 structurally representative TCM components in a simulated decoction environment. Furthermore, quantitative structure-property relationship (QSPR) models correlating structure and self-assembly binding energy were constructed using multiple machine learning (ML) algorithms. Isothermal titration calorimetry (ITC) was utilized to construct an independent external test set, strictly validating the external predictive performance of the established QSPR models on unseen compounds. Finally, cluster analysis was applied to systematically explore the structural characteristics of components capable of self-assembly. Kinetic and thermodynamic evaluations revealed that at an optimal concentration of 10 mM, 46 components show a favorable assembly trend, while exactly 28 components could spontaneously form stable supramolecular structures (ΔG < 0) primarily driven by non-covalent interactions (e.g., π-π stacking and hydrogen bonding) closely related to their specific structural features. Among the ML algorithms tested, the model built using the ExtraTrees algorithm performed best (R² = 0.963, RMSEP = 3.075), proving suitable for predicting the self-assembly binding energy of TCM components. Analysis of descriptor importance in the model and clustering results revealed that molecules with smaller molecular weight (Mw < 326 Da), a higher number of carbonyl groups connected to aromatic rings (nArCO ≥2), and greater molecular planarity were more prone to form supramolecular assemblies in the simulated decoction environment. The present work provides a rapid and reliable computational approach for screening monomer components capable of forming supramolecular assemblies in TCM decoctions and lays a methodological foundation for the efficient development of supramolecular nanodrugs and nano-delivery systems based on TCM decoctions system.