Machine Learning

We integrate quantum chemical topology with machine learning to build physically motivated neural network interatomic potentials and to accelerate the discovery of new chemical bonding descriptors.

A tool for the automated generation of optimised Atom-Centred Symmetry Functions (ACSFs) for neural network interatomic potentials, using Gaussian Mixture Models to characterise the chemical space of a system.

A Python-interfaced neural network model for the rapid prediction of QTAIM atomic charges of C, H, O and N atoms in gas-phase organic and biological molecules.

A deep learning code based on the SchNet architecture for training models on atomic (1-body) and pairwise (2-body) QTAIM properties. Supports CPU and GPU execution.

A graphical user interface for the prediction and visualisation of QTAIM atomic properties using feed-forward neural network models. Includes the built-in NNAIMQ model for Bader charges.