scientific-deep-chemistry

What problem does it solve?

Predict molecular properties from molecular structures using deep learning, enabling faster screening and design in cheminformatics.

Core Features & Use Cases

• Graph convolution and message passing models (GCN/MPNN/AttentiveFP) for property prediction on molecular graphs.
• MoleculeNet benchmark support and evaluation to compare model performance.
• Pretrained representations and embeddings (ChemBERTa, GROVER) for downstream tasks and transfer learning.
• End-to-end pipelines from data loading to prediction, with modular components for experimentation.

Quick Start

Train a GraphConv model on a MoleculeNet dataset to predict a molecular property.