The code in this repository implements an analysis pipeline for tissue-specific protein-protein interaction networks.

In this text we use the following abbreviations:

• PP: protein-protein interaction network
• tsPPI: tissue-specific PPI

Different parts of the pipeline are accomplished by different programming languages. Python is used mainly for data import and processing. Most data comes in text format (tab separated or comma separated) from various sources. This data is imported and processed into a unified format and saved into a SQLite table.

Graph analysis is done with the NetworKit parallel graph analysis library in order to speed up graph analysis tasks. To this end we implement our own extensions and extended algorithms for analyzing multiple Subgraphs simultaneously. This is useful, since all tsPPIs are represented as a set of subgraphs or a common PPI parent network. We implement this Subgraphs data structure and the extended algorithms using C++ and make them accessible inside our python based pipeline by extending NetworKit's cython API.

Final data analysis and visualization is done by a set of R scripts, making heavy use of the ggplot2 visualizations library.

The code is organized into different parts:

• src contains the python pipeline and other python scripts
• src/pappi contains all custom python modules used for the python pipeline
• analysis contains R scripts for analyzing data and generating figures and plots
• figs contains generated figures which are saved by the R scripts in analysis
• data contains the raw data and scripts to automatically download it
• TODO: add other repos (ppi_networkit, networkit and fastSemSim) and their dependencies

• python3 (sqlite3)
• NetworKit
• C++11 compiler
• fastSemSim

TODO (building python modules

Most of the datasets can can be automatically downloaded. Only a few need to be manually procured. See the data/README.md for details.

Make sure all needed data is available in the data folder before proceeding.

Running the python pipeline consists of running 3 steps: (approximate run times are on a 4 core 3.4Ghz Intel CPU with 8 GiB RAM)

1. Data import and preprocessing (~10-20 minutes)
2. Running the graph analysis on all networks (~4-8 hours)
3. Running graph clustering/community-detection on all graphs (~6 hours)

To run this pipeline, execute the following:

cd src
python3 init_data.py
python3 graph_properties.py
python3 networkit_clustering.py

The final analysis and data visualization is implemented in R, all scripts are available in the analysis folder. Run these scripts to get the individual figures and graphs.

The different BPScore algorithms can be benchmarked using the bpscore_benchmark.py script. To run this execute:

cd src
python3 bpscore_benchmark.py

executing benchmarks and tests of ppi_networkit