drGAT
Attention-based GNN model for predicting drug sensitivity and explaining gene contributions using a drug-cell-gene heterogeneous network.
Overview
drGT is a deep learning model that predicts drug response (sensitivity) in cancer cell lines while providing gene-level explanations through attention mechanisms. It constructs a heterogeneous graph connecting drugs, cells, and genes, enabling both accurate predictions and interpretable insights into which genes influence drug efficacy.
Key Features
- Attention-based GNNs — Utilizes GAT, GATv2, or Transformer architectures to model complex relationships in the drug-cell-gene network.
- Drug Response Prediction — Predicts drug sensitivity (binary or continuous) across cell lines and drugs.
- Gene Explainability — Uncovers gene-level contributions to predictions via attention scores, aiding biological interpretation.
- Pretrained Models — Includes pretrained models for inference without retraining on supported datasets.
- Flexible Training — Supports custom datasets, various data-splitting strategies (random, leave-drug/cell-out), and architecture modifications.
- GPU Acceleration — Optimized for GPU execution via PyTorch and PyTorch Geometric for faster training and inference.
Philosophy
drGT emphasizes interpretability alongside predictive accuracy, using attention mechanisms to bridge deep learning with actionable biological insights in drug discovery.
Related Projects
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
