Scripts, data, and supplementary information for the publication:

Graph algorithms for predicting subcellular localization at the pathway level. Chris S Magnano, Anthony Gitter. Pacific Symposium on Biocomputing, 2023.

Supplementary text, figures, and tables can be found in supplement.pdf

Repository overview

• caseStudy - Code and data used for the case study predicting subcellular localizations in primary fibroblasts during human cytomeglovirus infection. Instructions for running this analysis are in the subdirectory.
• data - Pathway data used for all experiments and scripts originally used to process that data.
• databasePrediction - Code and data used for predicting pathway database subcellular localizations from protein-level subcellular localization data. Instructions for running this analysis via Snakemake are below.
• exploratoryAnalyses - Preliminary figures from developing localization models and exploring possible pipelines using approximately 100 Reactome pathways in the 'Developmental' category.

Setting up the environment

1. Install Anaconda if needed.

   The easiest way to install Python and the required packages is with Anaconda. The Carpentries Anaconda installation instructions provide guides and videos on how to install Anaconda for your operating system.

2. Create the conda environment.

   From the pathway-localization directory, run:

   conda env create -f environment.yml
   conda activate pw_loc
   

   to create a conda environment with the required packages and activate that environment.

Recreating plots on archived results

The files in the databasePrediction directory can be used to recreate plots from the manuscript. First, extract the file in allRes.tar into the results directory:

tar -xvzf allRes.tar

Then run:

python plotResults.py results/allRes.p

This script will interactively display all plots from the database prediction experiment.

Running localization prediction

After following the steps in Setting up the environment, navigating to the directory databasePrediction and using the command:

snakemake --cores [n] all

will run the full localization prediction experiment, where [n] is the number of desired cores.

This workflow ends by running plotResults.py, which recreates the database prediction figures from the paper interactively. It does not save any figures automatically.

The file results/allRes.p contains 2 Pandas DataFrames saved using PyTorch, which store results metrics for each pathway and model configuration. metrics contains the per-pathway metrics for each run and pathway, while mergedMetrics contains metrics calculated by considering all edges at once to perform a single metric calculation per model. They can be loaded into Python as:

import torch
data = torch.load('results/allRes.p')
metricDF = data['metrics']
metricMergedDF = data['mergedMetrics']

Datasets

• Reactome data were from version 72 retrieved on June 26, 2020. These files were licensed with the CC0 license.
• Pathway Commons data were from version 12 retrieved on June 26, 2020. These files have the same licenses as the original source databases. The PathBank files were licensed with the ODbL v1.0 license.
• ComPPI data were retrieved on November 9, 2022. These files were licensed with the CC BY-SA 4.0 license.
• Compartments data were retrieved on September 29, 2021. These files were licensed with the CC BY 4.0 license.
• UniProt data were retreieved in October 2021. These files were licensed with the CC BY 4.0 license.

Any publications using these datasets should cite the original sources of the data.

Licenses

All software (e.g. files matching the patterns Snakefile, CMake*, *.py, *.h, *.cpp, and *.sh) is available under a MIT license (LICENSE-MIT.txt).

All other files (including data, figures, and the supplementary information) are available under a CC BY 4.0 license (LICENSE-CC-BY.txt) except for those derived from PathBank and ComPPI. Files derived from PathBank (data/labeledPathBank.zip) are available under a ODbL v1.0 license in accordance with the PathBank terms of use. Files derived from ComPPI (databasePrediction/data/comPPINodes.tsv) are available under a CC BY-SA 4.0 license in accordance with the ComPPI terms of use.