MC EM CS NEM

This repository serves as

• The location for hosting the supplementary materials for the paper "Context-Specific Nested Effects Models" by Sverchkov et al., to appear at RECOMB 2018, these files are in the recomb-2018-supplement folder.
• The source code repository for the simulation studies in the paper.

Files and where to find them

• recomb-2018-supplement contains a PDF of supplementary text and cytoscape files of the yeast salt stress network.
• R contains R code for assembling the result summary csv as well as R code for running the simulation and learning
• csv holds csv files, notably including the summary table.
• rdata would be created by simulation code, and hold RData files, including the ground truth generating models+data, learned models, evaluation statistics.
• plots would be created by plotting R scripts
• local-exec contain bash scripts for running the simulations
• json the simulated ground truths, in json

File naming conventions

Files created in rdata follow the pattern {type}-r{rep}-n{number of actions}-e{number of effects}-d{edge density}-k{true k}-b{beta parameter}-l{learning k} where (type) is truth/data/model, data doesn't have a learning k, and truth has neither a learning k nor a beta parameter

Simulation workflow

• Each of run_recomb2018.sh, do_density_runs.sh, or do_noise_runs.sh in local-exec creates ground truth and data if they do not exists, and learns models. One can first run generate_data.sh to ensure all models are created first. (Note that all of this takes weeks to run on a single machine.)
• The R script result-table-from-models.R reads the learned models and creates a csv file listing them and their precision/recall on effect matrix recovery and ancestry recovery.
• The R script make-plots.R makes the plots summarizing this.

Note on nem package

I use my own fork of the NEM package, for running MC-EMiNEM more memory-efficiently, and for access to some functions that the Bioconductor package doesn't expose.