Accurately determining tautomer ratios in solution poses a considerable challenge. Classical molecular mechanics force fields cannot describe the nuanced topological variations inherent in tautomerism. In contrast, neural network potentials (NNPs), trained on quantum mechanics data, hold promise for more adeptly computing tautomer ratios.
Here, we use relative free energy calculations with different neural network potentials to calculate tautomer ratios. We assess the accuracy of the tested neural network potentials on a subset of the Tautobase, which encompasses experimentally determined free energy differences for various tautomer pairs.
We are able to show that it is trivial to calculate tautomer ratios with neural network potentials and that the relative speed of the latest generation of hardware and neural network potentials allows for a correct treatment and setup of these calculations.
Sara Tkaczyk