pan-ontology-ml

This repository holds an analysis of the landscape of ontologies used in biomedical research. The analysis considers relationships between ontology terms defined through ontology hierarchies and through implicit semantic similarity measures as captured by machine-learning models.

Setup

This repository can be used to perform an analysis from scratch, or in conjunction with a prepared dataset.

Analysis from scratch

After cloning the analysis repository, set up the following files and directories.

crossmap script - bash script that executes the crossmap program.
crossprep script - bash script that executes the crossprep python utility.
data directory - a directory that will hold data files and processed items.
a mongodb database compatible with crossmap

With the software and database in place, the next phases are to download ontology datasets and run analyses. These steps are described in the README in the scripts directory.

The whole procedure utilizes more than 200 ontologies and performs several calculations on each ontology. The total running time may well exceed 100 hours.

Analysis using prepared data

A snapshot of required datasets is available at zenodo. Download the snapshot zip file into the repository root and uncompress it. That should create a directory data with all raw and processed files.

Visualizations

Visualizations are achieved via rmarkdown vignettes. To create these, navigate into the vignettes directory, launch R, and render the vignettes.

library(rmarkdown)
render("OntoML.Rmd")        
render("OntoML_Supplementary.Rmd")

During the first rendering, the vignettes will generate several files that will be stored under vignettes/cache. The first render will also require several minutes of compute and have a moderate memory footprint (16GB of RAM). Subsequent renders will be faster and more frugal with memory.

Repository structure