Software Development
MXtalTools
As part of a strategy for developing machine learning tools for molecular crystal polymorphs, DMREF investigator Michael Kilgour has introduced MXtalTools (Molecular Crystals Tools), a Python toolkit under ongoing development for machine learning tasks on molecular crystals. It includes utilities for dataset collation/curation, crystal modelling (regression, classification, generation), reporting, and a fast & differentiable module for crystal parameterization and construction. Find it at GitHub here.
MXtalTools (Molecular Crystal Tools) https://github.com/InfluenceFunctional/MXtalTools
Image courtesy of Michael Kilgour
MXtalTools — Classification
Extending graph-based neural networks to the problem of characterization of local molecular environments, this branch of the MXtalTools suite provides tools for dataset preparation, training, and analysis of bulk periodic or finite-size molecular structures using powerful graph models. Find it at GitHub here.
MXtalTools (Molecular Crystal Tools) — Classification https://github.com/InfluenceFunctional/MXtalTools/tree/mol_classifier
Image courtesy of Michael Kilgour
Automated Molecular Dynamics Preparation
An end-to-end automated scripting and configuration package for preparing molecular dynamics simulations using the LAMMPS code for nicotinamide as a bulk material, finite clusters in vacuum, or nanocrystalline fragments in the melt. Allows for detailed control of system size, defects, and other simulation properties from a convenient interface. Find it at GitHub here.
Automated Simulation Preparation for LAMMPS https://github.com/InfluenceFunctional/LAMMPS/tree/master
Image courtesy of Michael Kilgour
CrystalMath
In an effort to accelerate and simplify the process of Crystal Structure Prediction, we are developing CrystalMath, a Topological/Geometrical CSP protocol, that performs a deep analysis on a large subset of molecular crystal structures in the Cambridge Structural Database and exploits the extracted data to derive mathematical principles of CSP and initiate a workflow that removes the necessity of developing system specific interaction models for the calculation of the lattice energy. Find it at GitHub here.
CrystalMath
Image courtesy of Nikolaos Galanakis
MolStrucClassifier
Classification of local polymorphic structures in molecular materials. This approach combines handcrafted features of the local environment with a multi-layer perceptron for classification. The features are derived from molecular symmetry functions, combining a point-vector representation with symmetry functions. Find it at GitHub here.
MolStrucClassifier
Image courtesy of Daisuke Kuroshima