An Integrated Framework to Predict the Strength of New SMILE Using Graph Attention Network

Abstract

Machine Learning (ML) and Deep Learning (DL) have significantly advanced various fields, including healthcare, finance, autonomous systems, and scientific research. In healthcare, these technologies have been widely applied to disease prediction, such as cancer and diabetes. However, the challenge of drug resistance persists, creating a need for more effective drugs. Developing new drugs is a complex, expensive, and time-intensive process, requiring innovative approaches to enhance efficiency. To address this, GAT-PDE (Graph Attention Network-Based Framework for Predicting Drug Efficacy) is proposed to predict the efficacy of the new drugs/SMILES for specific diseases. The framework incorporates Pharmacophore fingerprints, Jaccard coefficient, quartile analysis, and Graph Attention Networks (GATs) to improve drug efficacy predictions. The Jaccard coefficient assesses molecular similarity between a reference drug and a database of one million compounds using pharmacophore fingerprints. Avapritinib, a proven drug for gastrointestinal stromal tumours (GIST), serves as the reference compound. Quartile analysis categorizes molecules based on Jaccard coefficient, generating labelled data. A GAT model is trained on this data, achieving 88% accuracy in predicting drug efficacy, demonstrating its potential for predicting efficacy of a new drug.