Fast and Accurate Prediction of Tautomer Ratios in Aqueous Solution via a Siamese Neural Network

Tautomerization plays a critical role in many chemical and biological processes, impacting the molecular stability, reactivity, biological activity, and ADME-Tox properties. Many drug-like molecules exist in multiple tautomeric states in aqueous solutions and complicating drug discovery. Predicting these tautomeric ratios and identifying the predominant species rapidly and accurately is crucial for computational drug discovery. In this study, we introduce sPhysNet-Taut, a deep learning model fine-tuned with experimental data leveraging the Siamese network. This model predicts tautomer ratios in aqueous solution using MMFF94-optimized geometries directly. On an experimental test set, sPhysNet-Taut surpasses all other methods, achieving state-of-the-art performance with an RMSE of 1.9 kcal/mol on the 100-tautomers set and an RMSE of 1.0 kcal/mol on the SAMPL2 challenge, and providing the best ranking power for tautomer pairs. Additionally, our results demonstrate that fine-tuning on experimental data significantly improves model performance compared to training from scratch. This work not only provides a useful deep learning model for predicting tautomer ratios, but also provides a protocol for modeling pairwise data. To facilitate user-friendliness, we developed a readily accessible tool to predict stable tautomeric states in aqueous solutions, enumerating all possible tautomeric states and ranking them using the sPhysNet-Taut model.