Code for inferring gene coexpression networks and phylogeny trees

• phylo_tree.py contains the PhyloTreeSample class, which generates and samples data from phylogenetic trees, and the PhyloTreeFit class which infers phylogenetic trees and co-expression networks from tissue gene expression data.
• unionfind.py contains the UnionFind class, an implementation of a disjoint sets data structure. From https://www.ics.uci.edu/~eppstein/PADS/UnionFind.py
• tester.py contains a script to generate phylogenetic trees, fit our algorithm and the neighbor joining algorithm, and compare the scores
• figures.py contains code to create the figures used in our presentation and report

To use the PhyloTreeSample class do the following:

//create original tree
pt = phylo_tree.PhyloTreeSample()
pt.generate_tree(.55) //a node will be the parent of other nodes with given probability 0.55
#create sampled tree
dat = pt.sample() //samples data from leaf nodes

Given a list of numpy arrays dat, you can fit a phylogenetic tree and co-expression networks with

pt_fit = phylo_tree.PhyloTreeFit()
pt_fit.fit(dat)

tester.py takes in 2 command line variables, the name of the file to write to and the number of trials to run. Usage:python tester.py filename n_trials The resulting csv file contains the following information for each trial:

• n_tissues: the number of tissues this trial
• tree_score: the value of the tree dissimilarity between the fitted tree and the real tree sampled (smaller is better)
• nj_score: the value of the tree dissimilarity between a tree fit using the neighbor joining algorithm and the real tree sampled
• cov_score: the mean square error of the co-expression matrices using the algorithm compared to the real co-expression matrices
• sample_cov_score: the mean square error of the sample covariance for each tissue compared to the real co-expression matrices
• time: the time to run the algorithm

This uses Python 3.4, and requires NumPy, SciPy, and Scikit-Learn.