QSAR/QSPR with GNNs

This project explores various Graph Neural Networks Algorithms for various machine learning (ML) tasks. Moreover, the underlying library provides many ways to represent chemical information, and explain the predictions. While the projects mostly applies techniques to solve small/large molecule ML, the library can be used for tasks on other types of graphs (e.g. Topic Classification for scientific articles, etc.).

Covered topics include, among others:

Tasks

1. Graph-level Classification
2. Node-level Classification
3. Link Classification - Coming Soon
4. Graph Regression - Coming Soon
5. Multi-task Learning - Coming Soon

GNN Algorithms

1. Graph Convolutional Networks
2. Graph Attention Networks
3. Graph Isomorphism Networks
4. MPNNs - Coming Soon

Sampling/Loading

1. Data Loading
2. Neighbor Loading

Explainability

1. GNNExplainer, SubgraphX, Gradient-Based Feature Attribution

Representation Learning, and Visualization

1. Representation Learning - Coming Soon
2. Molecular Graph-based Embedding, and visualization

Clustering

1. Scaffold Splitting
2. Cluster-Based Splitting

Visualizing Training Progress and Metrics

1. Tensorboard