README.md

# AIIM2023
<img src="./logo/logo.svg">

## Content in each directory:


- **documentation**: Instructions to install the needed libraries.
- **graphData**: Data of DISNET's graph.
- **metrics**: Training, testing and RepoDB verification ROC & PRC, also loss evolution for training and validation.
- **models**: Trained models.
- **results**: Result files of the RepoDB test and the distribution plots (once these are generated).
- **Code files:**
  - deepSnapPred (train & test DeepSnap) 
  - heterograph_construction (build graph)
  - testRepoDB (validate model using RepoDB) 
  - topN (getting top N new predictions of a model)
  - utilities (plotting utilities)
  - visualizeDistribution (Checks the distribution of the predictions for a group of diseases, check TFG for more information).
  - visualizeEmbeddings (visualization of embeddings)

## Summary
Using Graph Neural Networks (GNNs) to predict new drug repurposing hypothesis. The objective is to predict new edges between 
drugs and diseases (edge inference) or predict the type of edge (drug) between two diseases (edge classification).

One models is generated using DeepSnap library. The model has an encoder-decoder (GraphSAGE-Dot product) architecture,
the use of different architectures will be studied in the future. This is the generated models and its status:
- 🟢 **DeepSnap:** Passed verification tests, proved to be functional and works well.

### Symbol legend:
- 🟢 : Ready to use.
- 🟡 : Use is not recommended due to major issues.
- 🟠 : Not usable