Probing Curcumin Reactive Conformers in Keto-enol Tautomerization Enhanced by Clustering with t-SNE

Abstract The extensive conformational space of flexible molecules poses a significant challenge for predicting chemical reactivity through quantum chemical methods. For curcumin, whose keto–enol tautomerization is crucial to its biological activity and associated with its therapeutic potential in conditions such as Alzheimer’s disease, selecting meaningful conformers is particularly challenging. Traditional strategies for conformer selection that rely on variations in torsion angles may result in an excessive number of conformers. Our previously developed fragmentation-based strategy addresses this issue but may overlook critical conformers in highly flexible molecules. In this study, we developed an integrated workflow that incorporates xTB metadynamics for comprehensive conformer sampling, Coulomb matrix descriptors with t-SNE dimensionality reduction for structural encoding, clustering to identify representative structures, and DFT validation of ground and transition states. This workflow yields a diverse yet manageable set of conformers, enabling a refined energy profile of curcumin tautomerization with reduced activation energy, thereby aligning the results more closely with the experimental values than our previous estimates. Beyond curcumin, our findings demonstrate that the integration of machine-learning-assisted clustering with electronic structure calculations provides an efficient and transferable strategy for capturing conformational diversity in flexible molecular systems.