etowahadams / GNN-molecular-classification

Testing data efficiencies of graph neural networks on molecular graph classification

Graph neural networks for molecular graph classifcation

This is a respository to experiment with the data efficiency of various common GNN models (such as GCN, GAT, GraphSAGE, and GIN) on molecular graph classfication tasks. It utilizes the GraphGym platform to run and evaluate models.

Datasets

So far experiments have been conducted using three different molecular graph datasets from the Open Graph Benchmark (OBG) collection. (1) ogbg-molhiv contains 41K molecules which are labeled according to their ability to inhibit HIV replication. (2) ogbg-molbace contains molecules which are labeled according to their ability to inhibit human beta-secretase 1. (3)ogbg-molbbbp contains molecules which are labeled according to their ability to cross the blood-brain barrier.

Configuration

Experiment configurations are stored in the /run directory.