(Coulomb) Local potential energy density–supramolecular energy (LPED–SME) machine learning prediction—a web application to obtain the local SME from simple inputs

We developed a freely accessible web application ( https://clfirme.pythonanywhere.com/ ) that uses supervised machine learning (ML) to predict (Coulomb) local potential energy density (LPED) for intermolecular and intramolecular interactions. The user-friendly interface accepts simple inputs: atomic charges of interacting atoms and interatomic distances, avoiding the complex calculations typically required by Quantum Theory of Atoms in Molecules topology. Our learning curves demonstrate model stability relative to dataset size, leveraging uncoded, smooth features with strong physical relationships to the target LPED. After testing six different ML models, we found that Regularized Gradient Boosting performed best, achieving excellent predictive capacity in both primary and secondary testing. This model has an average uncertainty estimate of 40.9%, with larger errors observed for highly negative LPED values.