Abstract:In pharmaceutical research, the strategy of drug repurposing accelerates the development of new therapies while reducing R&D costs. Network pharmacology lays the theoretical groundwork for identifying new drug indications, and deep graph models have become essential for their precision in mapping complex biological networks. Our study introduces an advanced graph model that utilizes graph convolutional networks and tensor decomposition to effectively predict signed chemical-gene interactions. This model demonstrates superior predictive performance, especially in handling the polar relations in biological networks. Our research opens new avenues for drug discovery and repurposing, especially in understanding the mechanism of actions of drugs.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is that existing deep - graph models often overlook the polarity of edges in the network (i.e., whether the drug mechanism of action is stimulatory or inhibitory) when dealing with the prediction of chemical - gene interactions in biological networks. This results in insufficient accuracy of the models in predicting interaction relationships in complex biological networks. Specifically, the paper proposes an advanced deep - graph model that utilizes Graph Convolutional Networks (GCNs) and Tensor Decomposition to effectively predict signed chemical - gene interactions. This model particularly emphasizes the ability to handle polar relationships in biological networks, thereby improving the prediction performance, especially when dealing with network relationships with positive and negative polarities.
### Background and Problem Description of the Paper
1. **Importance of Drug Repurposing**
- The drug repurposing strategy can significantly reduce the R & D cost while accelerating the development of new therapies.
- For example, Doxycycline was rapidly applied in the clinic through drug repurposing during the COVID - 19 pandemic, improving the efficiency of public health response.
- Thalidomide was once banned due to teratogenic effects, but was later repurposed to treat multiple myeloma and other cancers, reducing drug development costs and increasing patient survival rates.
2. **Role of Network Pharmacology**
- Network pharmacology provides a theoretical basis for discovering new drug indications by constructing and analyzing the interaction networks between drugs and their biological targets.
- For example, by analyzing the interactions between known drugs and disease - related targets through network pharmacology methods, potential new indications of drugs can be discovered.
3. **Limitations of Existing Models**
- Existing deep - graph models usually ignore the polarity of edges (such as activation or inhibition) when dealing with biological networks, which leads to insufficient accuracy of the models in predicting complex interaction relationships.
- For example, in the DrugBank database, the relationships between drugs and targets can be polar, such as "agonist", "activator", "inducer" representing positive effects, while "inhibitor", "blocker", "antagonist" representing negative effects. These subtle biological meanings are often overlooked in unsigned networks.
### Solutions
1. **Model Design**
- An advanced deep - graph model that can handle signed networks is proposed.
- The model not only overcomes the limitation of existing methods in ignoring network polarity but is also compatible with non - polar relationships.
- By identifying complex multi - faceted relationships, the model provides a more detailed and dynamic perspective on chemical - gene interactions.
2. **Experimental Verification**
- The performance of the model was verified through multiple evaluation metrics (such as AUROC, AUPRC, AP@20).
- The experimental results show that the Paverage prediction model performs best on all three metrics, with an AUROC of 0.966, an AUPRC of 0.961, and an AP@20 of 0.978.
3. **Conflicting - Edge Sampling Strategy**
- The Cannot - Link (CL) conflicting - edge sampling strategy was introduced to enhance the model's discrimination ability when dealing with polar data.
- The experimental results show that after using the CL sampling strategy, the accuracy of the model in dealing with polar relationships is significantly improved, although the performance on some traditional metrics is slightly decreased.
4. **Influence of Sub - graph Structures**
- The influence of introducing chemical - chemical sub - graphs and gene - gene sub - graphs on the model's prediction performance was explored.
- The results show that introducing sub - graph structures may lead to a decline in model performance, so simplifying the model structure may be more advantageous on specific datasets.
### Conclusions
This research significantly improves the prediction performance of signed chemical - gene interactions by introducing an advanced deep - graph model and a conflicting - edge sampling strategy. This provides new avenues for drug repurposing and new drug discovery, especially in understanding drug mechanisms of action. Future work will further explore the relationship between model complexity and prediction performance, and how to adjust the model structure to maximize prediction accuracy.