roohy / localIBDClustering

Local IBD Clustering

This repository is created provide public access to the software material for our paper "Selecting clustering algorithms for IBD mapping".

Supplementary Figures and Methods are available in the sca_supmat.pdf file.

• run_sim.py

  In this file, we have added the code to conduct a single simulation and analysis as an example.

• igraph_reader.py

  The script for reading weighted edge list files into iGraph for further analysis and visualization

• igraph_pval.py

  Given a graph clustering from iGraph and original ground truth cluster array, this file calculate the statistical power of the clustering algorithm in a simulation.

• hcs2.py

  In this file, we have update the HCS algorithm implementation from this repo to support the latest version of NetworkX python library. We have also added the ability to analyze disconnected graphs to the algorithm which is essential for IBD clustering applications.

• clustering_tests.py

  This file includes the script that enable the calculation of clustering scores such as NMI and statistical power in the real_data_analyzer.py file.

• real_data_analyzer.py

  This is the main file that include most of the logic of the simulations. This file being with a set of functions that are in charge of converting the various graph formats such as tuple_2_vertex function that converts the output of MCL clustering library to iGraph partition objects.

  • Local IBD Graph Class: This class reads a single local IBD graph file and analyzes it using the selection clustering algorithms in the partition function. The algorithms currently include: MCL, Infomap, Louvain, Leiden, and HCS. It records the time it takes each algorithm to analyze the graph. calculate_stats function is in charge of recording clustering metrics for each algorithm based on its performance. When no ground truth is available, this class only calculate structural metrics.

  • Local IBD Generator Class: This class generetes local IBD graph files using the benchmark procedure layed out in our paper.