Codes and data for the paper entitled "Learning representations of molecules to predict intermolecular interactions by constructing a large-scale heterogeneous molecular association network"

contains 8 kinds of molecules, 18 types of associations, 14,315 nodes, 114,150 molecular associations.

The associations among miRNA, mRNA, circRNA, lncRNA, protein, drug, disease, and microbe, including miRNA-disease associations, circRNA-disease associations, circRNA-miRNA associations, disease-mRNA associations, disease-microbe associations, drug-disease interactions, drug-mRNA associations, drug-microbe associations, drug-protein interactions, lncRNA-disease associations, lncRNA-mRNA associations, lncRNA-miRNA associations, lncRNA-protein interactions, miRNA-drug associations, miRNA-mRNA associations, miRNA-protein interactions, mRNA-protein associations, protein-protein interactions.

Read the original association to construct the MAN network, and number the nodes

python MainFirst.py 

Randomly select the same number of negative samples as the positive samples from the association matrix, and and divide all samples into 5-fold for subsequent cross-validation.

python MainSecond.py

Obtain the network behavior features (network embedding) of biomolecule nodes in the MAN network

python SecondOpenNE-master/prepareFea.py

Generate attribute features, network behavior features, and attribute + network behavior features required for cross-validation

python MainThird.py

(The codes are in the "Train5Fold" directory)

The performance of attribute + behavior(manner) features

python attribute+manner.py

The performance of attribute feature

python attribute.py

The performance of network behavior feature

python manner.py

Classifier comparison (The codes are in the "Train5Fold\4-ClassifierCompare" directory)

# Run the following command in the folder of each classifier
python attribute+manner.py 

Code and data are in the "case study\3TrainTestAttribute+Embedding" directory

python attribute+manner_casestudy.py

This can be regarded as a link prediction problem on MAN. Evaluating the performance of the proposed method for a specific interaction (e.g., a protein-protein interactions) can be done as follows.

1. Remove the protein-protein interaction pairs (links) from molecular association network as training set. 
(Just comment out the csv file name  corresponding to the type of interaction to be tested (for example: PPI.csv) when reading all the associations to build the network.)

2. Training the proposed MMI-Pred model using training set.

3. Using protein-protein interaction pairs as the test set to test the performance of the model.

openne
numpy==1.14
networkx==2.0
scipy==0.19.1
tensorflow>=1.15.2
gensim==3.0.1
scikit-learn==0.19.0

Yi, H.-C. et al. Learning representations of molecules to predict intermolecular interactions by constructing a large-scale heterogeneous molecular association network. iScience, doi:10.1016/j.isci.2020.101261.

Contact: haichengyi#gmail.com