GraphDTA Replication with PyTorch: Implementation, Code, and Results

This technical report explores the mathematical foundations and implementation of a GraphDTA replication in PyTorch for drug–target affinity prediction. It reviews molecular graph construction from SMILES with RDKit, protein sequence encoding via 1D CNNs, and graph representation learning with GCN and GAT, followed by global pooling and a regression head. Training and evaluation are documented on the Davis (and optionally KIBA) datasets using MSE and the concordance index, alongside data preprocessing, batching, and reproducibility tests. The report consolidates methodology, theory, and results as part of an independent project in computational drug discovery and bioinformatics.