Deep Learning and Graph Neural Networks for Drug-Like Molecule Modeling

Abstract

The recent developments in artificial intelligence have had a significant influence on drug discovery, with the possibility of modeling drug-like molecules based on data. Graph neural networks and other deep learning methods have been shown to be effective for learning useful molecular representations, as well as for learning complex chemical structures directly based on data. The chapter offers an introduction to deep learning analysis through the focus on graph-based algorithms of the prediction of molecular properties, drug-likeness, and de novo generation of molecules. It talks about prominent molecular representations like SMILES, molecular fingerprints, and molecular graphs, and how they are incorporated into the current neural network models. The chapter also examines evaluation metrics and reports real-world research on the usefulness of these models in practical drug discovery. Lastly, it outlines the existing constraints, research gaps, and challenges in the future, insisting on the increased use of deep learning and GNNs in pharmaceutical discovery speeds and improvements.