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

Area of Science:

• Computational Chemistry
• Machine Learning
• Cheminformatics

Background:

• Molecular machine learning (ML) utilizes molecular descriptors or fingerprints for property prediction.
• Graph neural networks (GNNs) offer comparable or superior performance to traditional ML methods in molecular tasks.
• Existing GNN tools lack seamless integration with TensorFlow and Keras APIs.

Purpose of the Study:

• Introduce MolGraph, a novel GNN package tailored for TensorFlow/Keras compatibility.
• Facilitate the generation of molecular graphs for GNN input.
• Validate GNN performance in molecular ML tasks and data interpretation.

Main Methods:

• Developed the MolGraph package with a focus on TensorFlow/Keras API compatibility.
• Implemented a module for generating small molecular graphs.
• Benchmarked GNN performance using MoleculeNet datasets and chromatographic retention time data.

Main Results:

• GNNs implemented via MolGraph performed as expected in benchmarking.
• The GNN approach demonstrated utility in molecular identification.
• Interpretability of chromatographic retention time data was enhanced using GNNs.

Conclusions:

• MolGraph provides a compatible GNN modeling pipeline for molecular ML within TensorFlow/Keras.
• GNNs show promise for molecular identification and improving the interpretability of complex chemical data.
• The MolGraph package is publicly available for broader research application.