How to use LAPINE

All LAPINE code must be run in the Conda virtual environment built during the installation process. However, as an exception, the network embedding process must be executed in the BioNev virtual environment. Please execute the python files in the order described below.

1.Download data

Run data_downlaod.py in your virtual python environment (above 3.7.2) constructed by conda to download all necessary files from the public database.

(.venv) python data_download.py

2.Network construction

Run network.py to parse network data from protein-protein interactions from the STRING database, and chemical-protein interactions from the STITCH database.

(.venv) python network.py --threshold 700

parameters

  • threshold : This parameter determines the threshold for the reliability of information included in STRING and STITCH databases. Only interactions with a confidence score higher than the parameter is used in the analysis. Therefore, a high parameter value means the use of limited high-reliability information only. The range of parameters is from 0 to 1000.

3.Network embedding

Run the following command in the BioNEV environments (python 2.7) to generate the embedding of the constructed network.

(.venv) conda activate bionev
(bionev) bionev --input './result/network.edgelist' --output './result/emb.txt' --directed True --method 'node2vec' --dimension 128  --p 4 --q 0.5

parameters

  • input : path of input file (do not change)

  • output : path of output file (do not change)

  • directed : consider network as the directed network (do not change)

  • method : name of algorithm for the network embededing. The default value is ‘node2vec’ which is optimized by cross-validation. See the links at the bottom of this section for other options.

  • dimension : the dimensions of embedding for each node. The default value is 128 which is optimized by cross-validation. The recommended parameter range is 16 to 512.

  • p, q : two parameters that control the biased random-walk procedure. A high p value ensures that the walker is less likely to visit the just visited node. A high q value ensures that the walker is more likely to visit the node which is close to the node walker just passed. The default values of p, q are 4 and 0.5 which is optimized by cross-validation. The recommended parameter range is 1/4 to 4.

more details are available at https://github.com/xiangyue9607/BioNEV

4.ADR-protein relation prediction

Run model.py to predict the association score between ADRs and proteins.

(.venv) python network.py --regularization 0.1 --threshold 700

parameters

  • regularization : Inverse of regularization strength of logistic regression, smaller values specify stronger regularization; must be a positive float. The default value is 0.1 which is optimized by cross-validation. The recommended parameter range is 0.01 to 1.

  • threshold : This parameter determines the threshold for the reliability of information included in STRING and STITCH databases. It must be the same or larger value used in network.py. Larger values allow you to selectively obtain less relation scores between ADRs and proteins that appear only in interactions with high confidence scores.(recommended to use the same value used in network.py)

Output

  • The output file predictions.csv contains scores for the associated ADRs (columns) for each protein (row).

  • The association score ranges from 0 to 1.

  • ADR terms are mapped to Preferred Terms (PTs) suggested by MedDRA, and proteins are mapped to Ensembl protein ID.

ADR example : C0000727, C0000731 / Protein example : ENSP00000410269, ENSP00000379364