Leveraging Deep Learning and Molecular Representation for Drug Discovery

With the rise of technology and ethical concerns, traditional approaches like wet lab experimentation in drug discovery has come under scrutiny due to their high cost, time-consuming nature, and environmental impacts. As a result, there is a growing push towards more sustainable alternatives. The selection of potential drug candidates requires extensive biological testing in the lab, and the most substantial costs in drug development is associated with research and clinical trials of drug candidates that may never receive approval. This paper addresses the imperative need to optimize the drug development pipeline for sustainability by the integration of virtual screening methods in drug discovery. Virtual screening methods, leveraging computational approaches, have gained importance in drug development. This involves screening large compound libraries using deep learning techniques, particularly neural networks, for molecular representation learning, crucial in predicting various molecular properties. In this work, the graph-based approaches are studied to capture information from the SMILES representation of molecular compounds. These approaches, rooted in Graph Neural Networks (GNNs), prove to help identify key properties of molecules, resulting in a reduction of experimental trial and error and, subsequently minimizing waste. The findings in this work serve to emphasize the significant role GNNs can play in addressing the challenges of conventional drug discovery approaches. The experimental results obtained in this work proved to be promising.