GNN-Based Drug–Target Binding Affinity Prediction Using Molecular Graphs and Protein Sequences

Abstract

The rapid and cost-effective prediction of drug-target interactions (DTIs) is a critical challenge in computational drug discovery. This project presents a novel web-based system that predicts Drug Target Affinity (DTA) using Graph Neural Networks (GNNs) and amino acid sequence embeddings. The model represents drug molecules as molecular graphs derived from SMILES strings and proteins as encoded sequences. A custom GNN architecture processes graph-structured molecular data while a convolutional embedding layer extracts features from protein sequences. The integrated model predicts binding affinity scores, enabling interpretation of interaction strength. The system includes a user-friendly interface for submitting single or batch predictions, visualization of molecule structures, interactive affinity plots, and 3D binding structure rendering using RDKit and 3Dmol.js. This solution demonstrates a powerful and extensible platform for virtual screening, offering interpretability and speed in early-stage drug development pipelines.