Efficient Black-Box Prediction of Hydrogen-Bond-Acceptor Strength

Hydrogen-bond-acceptor strength is a critical determinant of physicochemical properties and binding affinity in drug discovery, but computationally predicting the strength of different hydrogen-bond acceptors remains challenging and inaccessible to non-experts. Here, we report a robust black-box workflow for predicting site-specific pKBHX values in organic molecules with minimal computational cost. Our approach begins with rapid conformer generation and optimization with neural network potentials, followed by a single density-functional-theory calculation of the electrostatic potential. We then calibrate these results against an extensive reference set of experimentally determined pKBHX data, achieving sub-0.2 pKBHX-unit accuracy. We illustrate the power of this tool in multiple published drug discovery programs, highlighting how per-site pKBHX tuning can improve bioavailability, minimize efflux, and enhance selectivity