Binding Affinity Prediction with Graph Neural Networks

• The repository contains the notebooks for the final chapter of my Master's Thesis.

• The notebooks contain the implementation of a novel fusion-based architecture, which combines a graph neural network (AttentiveFP) and biologically-motivated learned protein embeddings (ProtTransBert) for drug-target interaction prediction. The model architecture is outlined below:

The model is benchmarked on Davis and KIBA standard datasets, as well as on an in-house dataset of ~400K de novo generated molecules against 6 targets.