DeepSignalingFlow: an integrative and interpretable graph AI framework to uncover mechanisms of synergy (MoS) in drug combinations by mining multi-omic data

---

Complex signaling pathways are often implicated in drug resistance. Combining drugs to perturb multiple signaling targets can mitigate resistance, but understanding mechanisms of synergy (MoS) remains challenging. Large-scale multi-omic resources and experimentally measured drug-combination synergy scores (e.g., NCI ALMANAC, O’Neil, DrugComb, DrugCombDB) create new opportunities to study MoS systematically. DeepSignalingFlow models the directed signaling flow from disease-critical proteins (with multi-omic evidence) to candidate drug targets—an approach not used by prior methods. Across four synergy datasets, DeepSignalingFlow achieves state-of-the-art prediction and interprets MoS via core signaling flows, paving the way for precision combination therapy.

---

Contents

• 1. Model Architecture
• 2. Run DeepSignalingFlow
  • 2.1 Parse the datasets
  • 2.2 Train baselines
  • 2.3 Train DeepSignalingFlow
• 3. Results
• 4. Cell-line Specific Biomarkers
• 5. Result Validation
• Citation
• License

---

1. Model Architecture

---

2. Run DeepSignalingFlow

2.1 Parse the datasets

python load_data.py

2.2 Train baselines

python geo_tmain_gat.py
python geo_tmain_gcn.py
python geo_tmain_gformer.py
python geo_tmain_gin.py
python geo_tmain_mixhop.py

2.3 Train DeepSignalingFlow

python geo_tmain_webgnn.py

---

3. Results

---

4. Cell-line–Specific Biomarkers

---

5. Result Validation

• Shortest-path weight sums (< 4 hops) between two drug-target genes (Top-10 vs Bottom-10 combinations, NCI ALMANAC):

  

• Shortest-path weight sums (< 5 hops) between two drug-target genes (Top-10 vs Bottom-10 combinations, NCI ALMANAC):

  

---

Citation

If you use DeepSignalingFlow in your research, please cite:

@article{zhang2024using,
  title={Using DeepSignalingFlow to mine signaling flows interpreting mechanism of synergy of cocktails},
  author={Zhang, Heming and Chen, Yixin and Payne, Philip and Li, Fuhai},
  journal={npj Systems Biology and Applications},
  volume={10},
  number={1},
  pages={92},
  year={2024},
  publisher={Nature Publishing Group UK London}
}

---

License

MIT License. See LICENSE for details.