MolGraph: a Python package for the implementation of small molecular graphs and graph neural networks with TensorFlow and Keras

Molecular machine learning (ML) has proven important for tackling various molecular problems, including the prediction of protein-drug interactions and blood brain-barrier permeability. Since relatively recently, so-called graph neural networks (GNNs) have been implemented for molecular ML, showing comparable or superior performance to descriptor-based approaches. Although various tools and packages exist to apply GNNs for molecular ML, a new GNN package, named MolGraph (https://github.com/akensert/molgraph), was developed in this work with the motivation to create GNNs highly compatible with the TensorFlow and Keras application programming interface (API). As MolGraph focuses specifically and exclusively on molecular ML, a chemistry module was implemented to accommodate the generation of molecular graphs $\unicode{x2014}$ which could then be inputted to the GNNs for molecular ML. To validate the GNNs, they were benchmarked against the datasets of MoleculeNet, as well as three chromatographic retention time datasets. The results on these benchmarks show that the GNNs performed as expected. Additionally, the GNNs proved useful for molecular identification and improved interpretability of chromatographic retention data.