Paper Repository

This repository contains the code to implement all examples in our paper: AdaPT: An interactive procedure for multiple testing with side information. Note that the code in this repository does not exactly replicate the results and the figures in the paper because our R package adaptMT was developed after the acceptance of the paper. For readers who want to exactly reproduce the results, please check our older repository AdaPT. In contrast, the adaptPaper repository serves as a tutorial of adaptMT package and we recommend the readers to follow it, rather than the AdaPT repository, for future research.

Introduction

All real experiments (Section 5.1 and Section 5.3), except the gene/drug example with random orderings, can be done in a laptop. The simulations, as well as the gene/drug example with random orderings, should be ran on a cluster. All R files are listed in the folder R/. The bash files to submit jobs to the cluster are listed in the folder bash/ (note that this depends on your cluster and the bash file might need to be changed correspondingly. See the section "cluster jobs" below for details). The outputs and the plots have already been contained in the folder data/ and the folder figs/, respectively.

Installing the package

Run the following code in R

devtools::install_github("lihualei71/adaptMT")

R files

The folder R/ contains all R files:

• test.R gives a toy test that checks whether the functions work;
• exprs_laptop.R implements all experiments (Section 5.1 and Section 5.3) that can be done on a laptop;
• estrogen_random_cluster.R, simul1_cluster.R and simul2_cluster.R implement the tasks (Section 5.1 and Section 5.2) that must be done on a cluster;
• aggregate.R aggregates the outputs from the tasks ran on clusters;
• adapt_exprs_plot.R reproduces all figures in Section 5;
• all other files are helpers.

A toy test

Our experimental results take several hours. We put a test file test.R that contains a toy task to check whether the functions work. We recommend running it before replicating the results. If no error is thrown to the console, the other files should be implemented without problems. The test takes around 2 minutes. If any error is encountered, please report the issue here or contact me by email.

cd R
R CMD BATCH test.R
cd ..

Experiments on the laptop

exprs_laptop.R generates the results for

• gene/drug data, Section 5.1, with highly informative ordering (data/estrogen_high_res.RData), with moderately informative ordering (data/estrogen_mod_res.RData),
• Bottomly data, Section 5.3 (data/Bottomly_res.RData),
• airway data, Section 5.3 (data/airway_res.RData),
• pasilla data, Section 5.3 (data/pasilla_res.RData),
• SILAC data, Section 5.3 (data/proteomics_res.RData).

The task takes around several hours. Please be patient.

cd R
Rscript exprs_laptop.R 
cd ..

Experiments on the cluster

estrogen_random_cluster.R, simul1_cluster.R and simul2_cluster.R should be implemented on a cluster. Each file reads two environmental variables: times and seed. times specifies the number of repeated experiments and seed specifies the random seed. In the folder bash/, there are three .sh files corresponding to each task. Each .sh files generates 50 jobs with times = 2 and seed ranging from 0 to 49. The outputs should contain 50 .RData files for each job, which are not listed in this repository. Finally, aggregate.R merges all outputs and turn them into a single .RData file for each task, namely estrogen_random.RData, simul1.RData and simul2.RData.

Figures

adapt_exprs_plot.R generates all figures in Section 5.

cd R
R CMD BATCH adapt_exprs_plot.R 
cd ..