Thermodynamics-informed neural networks and extensive data sets: key factors to accurate blind predictions of apparent p<i>K</i>_a values in the euroSAMPL challenge

Microscopic and macroscopic p<i>K</i>a values for 35 compounds selected by the organizers of euroSAMPL 1 challenge were blindly predicted with our thermodynamics-informed empirical S + p<i>K</i>a model (ranked submission 0x4cb7101f). Our results have received the first overall rank from the challenge organizers. We describe our methodology and discuss evaluation methods.