DDNet: Drug Drug Interaction Prediction with Meta-Paths and Local Structures

This project is to predict drug drug interactions by an assumption says similar nodes tend to show similar behaviors. Therefore, we set a network and try to represent drugs in new dimensions and also we add topological information by concetenate meta-paths. This github project can be downloaded with

git clone https://github.com/atakanahmetce/g-DTIPred.git

Background

Representational learning is done via node2vec algorithm that uses only nodes' 1-hop or 2 hops neighbors information. The reason why not using entire graph is to decrease complexity and to increase node and graph influence. The way we applied is that:

Figure.1 An example of how subgraphs are generated and feature embeddings

• First, we generate subgraphs (1-hop and 2-hops) for each sample
• Second, subgraphs are given to node2vec algorithm (it is modified word2vec algorithm)
• Finally, new representaions are obtained in new dimension

Then, we concetenate meta-paths between nodes to add topological information of the drug network to node2vec feature of drugs. These paths are calculated according to length of the paths between two nodes. (Figure 2)

Figure.2 An example of meta-path features generation and feature vector reconstruction

Use Case

Project guidlines are given in use-cases.

• System: Operating system independent
  
• Required Packages: numpy-1.18.5, scipy-1.4.1, pandas-1.3.3, scikit-learn-1.0.1, networkx-2.5, pytorch-1.9.1, rdkit-2021.09.1, node2vec-0.4.5, tqdm-4.63.0

To install node2vec algorithm:

pip install node2vec

Data sources

• drug SMILES data: CROssBAR
• drug-drug interaction: DrugBank

PPI and DTI interaction fetched via CROssBAR Data API
https://www.ebi.ac.uk/Tools/crossbar/swagger-ui.html