GSchNetTools

A selection of tools for analysing ASE databases generated by G-SchNet, complete with a nice CLI for ease of use.

Features:

• Parsing of 'bonding' features from databases (number of elements, bonds, rings, aromaticity, etc.) and saving as new ASE databases with full metadata.
• Fitting of PCA models for dimensionality reduction of structural (SOAP/MBTR) and bonding (see above) descriptors.
• Transformation and export of principal components from fitted models, for use in chemical space plotting.
• Subsampling of databases to match the molecular weight distruibution of a training database.

Installation

Since the internals of this package depend on Open Babel, it's probably easiest to install through pip within a conda environment, as the openbabel package is much easier to install through conda.

While in a conda environment with openbabel already installed, clone the repository and then run

pip install --no-deps ./GSchNetTools

A better installation solution is coming, I just haven't got around to uploading to conda-forge yet.

Usage

Once installed, the command line program gstools will be available. This has a full-featured help system that can be accessed by calling gstools help {subcommand}, but to get you started:

• gstools parse [db] {options} parses a database for bonding features.
• gstools fit_pca [dbs] {options} fits PCA models on the molecules in one or more databases.
• gstools transform_pca [pca_models] [dbs] {options} uses the models from the previous command to fetch and save PCs for the descriptors of the molecules in one or more databases.
• gstools sample [train_db] [dbs_to_sample] {options} subsamples databases according to the molecular weight distribution in train_db.