About

MXtalTools is a Python library for machine learning on molecules and molecular crystals, built on PyTorch and PyTorch Geometric.

The library provides:

• Crystal building – Fast, differentiable construction of molecular crystal supercells from asymmetric unit parameters and space group symmetry.

• Crystal density prediction – Predict crystal packing coefficients from molecular structure using a pre-trained graph neural network.

• Molecule autoencoder – Encode molecules into equivariant vector and scalar representations using a pre-trained Mo3ENet model.

• Crystal scoring – Evaluate crystal structures against CSD statistics using a trained classifier.

• Crystal structure search – Optimize crystal packing parameters using machine-learned interatomic potentials and scoring models.

• Dataset utilities – Tools for constructing molecular and crystal datasets from CSD, .cif, and .xyz files.

• Model training – Configurable training workflows for graph neural networks on molecular crystal tasks.

Reference

If you use MXtalTools in a publication, please cite: