ExactL3 is a better normalized link predictor for Protein-Protein Interaction (PPI) Networks.

Published at IEEE-BIBIE 2020, awarded with the Best Bioinformatics Paper Award, link: https://doi.org/10.1109/BIBE50027.2020.00017

This repo includes the introduced algorithms, the experiments to generate all the results data, a jupyter notebook that formats all the result images, and a command-line script to run ExactL3 and various link predictors for PPI networks.

Language: Python. Python Libraries (to run the program): numpy, pandas

Workflow: example_PPI.txt (input file) => ExactL3_cmd.py => example_PPI_ExactL3.txt (output file)

• To run ExactL3_cmd.py in the terminal (assume Windows, cmd): python ExactL3_cmd.py {input file path} {output file path} {link predictor} {number of CPU core}

• Example to run ExactL3: python ExactL3_cmd.py ./example_PPI.txt ./example_PPI_ExactL3.txt ExactL3 1

• Supported Link Predictors (see the paper for reference / details): ExactL3, L3, CN, CRA, Sim, CH2_L3

The input file is a tab-delimited .csv file with no header, where there are two columns. The number of rows is the number of PPIs, and for each row the two columns of that row induce an non-directional PPI (each item is a protein). For example, a row A\tB implies protein 'A' and protein 'B' has a PPI.

For examples to work with our Python script, see ./example.py. Documentations are included as comments in the script.

The data in our paper is generated using the script ./src/notebook/dataGen_Yeast.py and dataGen_Human.py, and the images in our paper are generated based on our generated data using the jupyter notebook ./src/notebook/ppiLPred_BIBE2020_Yeast.ipynb and ppiLPred_BIBE2020_homo.ipynb. Note that except the parsed datasets in ./src/data/parsed/ and generated sample PPI datasets ./src/genData/, no data is included since it is too large (roughly 700 GB).

For more explanations how our python functions realize the algorithm, see here