This research project implements a novel deep learning architecture combining 3D Convolutional Neural Networks (CNNs) and Graph Convolutional Networks (GCN) to predict protein-ligand binding affinities. The model processes structural and chemical information through parallel networks before combining outputs for final prediction.
- Dual 3D-CNN branches processing voxelized molecular representations (48×48×48×19)
- Graph Convolutional Network (GCN) for molecular topology
- Multi-Layer Perceptron (MLP) for final affinity prediction
- Based on the PDBbind dataset
- Preprocessed molecular structures with computed charges (MOL2 format)
- Voxelized representations for CNN input
torch>=1.9.0
numpy
pandas
h5py
biopandas├── LICENSE
├── README.md
├── data
│ ├── pdb_bind
│ │ ├── refined-set # Refined Set Data
│ │ └── v2020-other-PL # General Set
│ └── sample_data
├── notebooks
└── src
├── models
├── preprocessing
└── training- to be added.
- to ber added.
MIT
[Your Name] [Your Institution] [Your Email]