Tautomerization plays a key factor in the natural biological process and drug discovery, yet it is often hard to obtain the ratio between two tautomers experimentally. Although the ratio depends on many factors, we looked for fast and robust methods to predict tautomer electronic energies. We used well-established DFT methods to calculate the energies of ~3000 pairs of tautomers from a tautomerization database. These results were compared with the ANI model (Accurate NeurAl networK engINe for Molecular Energies, or ANI). We used two ANI modes: ANI-1ccx and ANI-2x. Compared with DFT results, the mean of molecular 3D root-mean-square deviations (RMSD) from ANI is about 0.20 Å, indicating that the structures optimized by ANI and DFT are almost exact. Our results showed that ANI model can approach DFT accuracy with little computational cost for molecular energy calculation and geometry optimization.

Benchmarking study of tautomer energies calculated with ANI and DFT methods

ACS Fall 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.