Open source software for constructing the Allegro potential model based on E(3)-equivariant graph neural networks and using the potential model for molecular dynamics simulations. The code depends on NequIP and can be run in a similar manner. Allegro scales better than NequIP since it doesn’t rely on message passing and the architecture is strictly local with respect to atom-wise environments.
Software tool for constructing interatomic potentials based on nonlinear atomic cluster expansion. It requires the user to either prepare a fitting dataset based on pandas and ASE, or it can automatically extract data from VASP calculation results. The obtained potentials can be used for molecular dynamics simulations using LAMMPS, and it also provides the capability to calculate extrapolation grades for on-the-fly active learning.
BEEMs is a Bayesian optimization tool of Effective Models (BEEMs). In BEEMs, the quantum lattice model solver HΦ is used as a forward problem solver to compute the magnetisation curve based on the given Hamiltonian. The deviation between the obtained magnetisation curve and the target magnetisation curve is used as a cost function, and the Bayesian optimization library PHYSBO is used to propose the next candidate point of the Hamiltonian for searching the minimum cost function
Python code for a chemical database, PubChem. Users can search data in PubChem by compound name, structural information and so on. It is also possible to receive outputs as a Pandas DataFrame.
Open source Python package for data mining of materials. It can extract data from more than dozens of databases, perform preprocessing and visualization of extracted data. By combining machine-learning tools such as scikit-learn, users can build machine-learning models with descriptors created from the extracted data.
Python tool for automatic extraction of chemical substance information from literature. Based on natural language processing algorithms, it can extract substance names and related physical/chemical properties such as melting points and spectra from documents written in English.
An AI system for predicting protein conformation. It is possible to predict the three-dimensional structure (folding structure) of a protein from its primary sequence (amino acid sequence). It learns hundreds of thousands of protein structure databases and uses DeepMind-based deep learning techniques to predict the conformation of new proteins from their amino acid sequences.
An open-source library for machine learning. Various functions on machine learning/deep learning are implemented in this package. Using flexible user-friendly description, various types of networks from simple to complex ones can be implemented. GPGPU parallel computation based on CUDA is also supported.
An application for structure prediction based on the genetic algorithm. This application can predict the structure and composition of stable phase of crystals, molecules, atomic clusters, and so on by using first-principles calculation and molecular dynamics. This application implements interfaces with various programs such as VASP, LAMMPS, MOPAC, GULP, JDFTx, etc, and runs efficiently on parallel computing architectures.
An application for prediction of stable and metastable structures from a chemical composition. This application applies the revolutionary algorithm to structure prediction by using various external energy calculators (VASP, GULP, Quantum Espresso, CASTEP).