Reproducing the simulations and real data analysis reported in `Doubly robust and computationally efficient high-dimensional variable selection’

Abhinav Chakraborty, Jeffrey Zhang, Eugene Katsevich

This repository contains code to reproduce the analyses reported in the paper “Doubly robust and computationally efficient high-dimensional variable selection” (arXiv, 2024).

Dependencies

Ensure that your system has the required software dependencies.

• R version 4.2.2 or higher
• Nextflow version 23.10.1 or higher

Get started

First, clone the symcrt2-manuscript repository onto your machine.

git clone [email protected]:Katsevich-Lab/symcrt2-manuscript.git

We used a config file to increase the portability of our code across machines. Create a config file called .research_config in your home directory.

cd
touch ~/.research_config

Define the following variable within this file:

• LOCAL_SYMCRT2_DATA_DIR: the location of the directory in which to store simulation results.
• LOCAL_EXTERNAL_DATA_DIR: the location of the directory in which to store external real data.

The contents of the .research_config file should look like something along the following lines.

LOCAL_INTERNAL_DATA_DIR="/Users/jeffreyzhang/data/projects/"
LOCAL_SYMCRT2_DATA_DIR=$LOCAL_INTERNAL_DATA_DIR"symcrt2/"
LOCAL_EXTERNAL_DATA_DIR="/Users/jeffreyzhang/data/external/"

Next, create an .Rprofile file in your home directory (if you have not yet done so).

cd
touch .Rprofile

Add the following command to your .Rprofile.

.get_config_path <- function(dir_name) {
  cmd <- paste0("source ~/.research_config; echo $", dir_name)
  system(command = cmd, intern = TRUE)
}

Simulations

Download simulation results data

Next, we recommend downloading the simulation results data from Dropbox, so that you can reproduce the figures without having to rerun the simulations. The data are stored in .rds format. Download the results directory from here: Dropbox results repository and place the results directory into LOCAL_SYMCRT2_DATA_DIR/private. This can also be done using the following commands: First,

source ~/.research_config
cd $LOCAL_SYMCRT2_DATA_DIR"/private"
wget --max-redirect=20 -O download.zip https://www.dropbox.com/scl/fo/qz1ctahx7tn5i2barich9/ABYSILth4DWkAyzqCc9umkU?rlkey=7lp7suuzvd126vdc4jxexynv5&dl=1

Then, execute

 unzip -o download.zip

If you would like to rerun the simulations from scratch, do not download the results and instead follow the steps in the next section.

Run the Nextflow pipelines

Navigate to the symcrt2-manuscript directory. All scripts below must be executed from this directory.

Also, for the commands below, depending on the limits of your cluster, you may need to set the max_gb and max_hours parameters differently. The defaults are 7.5 and 4, respectively.

# tower PCM statistical simulation:
echo "bash code/run_simulation_pipeline.sh --sim_name split_pcm_stat_manu" | qsub -N run_all

# PCM statistical simulation:
echo "bash code/run_simulation_pipeline.sh --sim_name oat_pcm_stat_manu" | qsub -N run_all

# oracle GCM statistical simulation:
echo "bash code/run_simulation_pipeline.sh --sim_name gcm_stat_manu" | qsub -N run_all

# HRT statistical simulation:
echo "bash code/run_simulation_pipeline.sh --sim_name hrt_stat_manu" | qsub -N run_all

# tower PCM, PCM, oracle GCM computational simulations:
qsub code/run_all_simulations.sh 

# HRT computational simulations:
qsub code/run_hrt_simulation_100.sh 
qsub code/run_hrt_simulation_125.sh 
qsub code/run_hrt_simulation_150.sh 
qsub code/run_hrt_simulation_175.sh 
qsub code/run_hrt_simulation_200.sh 

Create the figures

Before creating the figures, please ensure that your working directory is set to symcrt2-manuscript. The figures are placed in the manuscript/figures directory.

# Figure 1
Rscript figures_code/plot_figure_1.R

# Figures 2,3,4.
Rscript figures_code/plot_stat_figures.R

# Choosing the splitting proportion (Figures 5-8 in the Appendix)
Rscript figures_code/plot_choose_proportions.R

Real data analysis

Download real data

Download the LIU22 directory from here: Dropbox LIU22 repository and place the LIU22 directory into LOCAL_EXTERNAL_DATA_DIR. This can also be done using the following commands:

First,

source ~/.research_config
cd $LOCAL_EXTERNAL_DATA_DIR
wget --max-redirect=20 -O download.zip https://www.dropbox.com/scl/fo/qz1ctahx7tn5i2barich9/ABYSILth4DWkAyzqCc9umkU?rlkey=7lp7suuzvd126vdc4jxexynv5&dl=1

Then, execute

 unzip -o download.zip

Navigate to the symcrt2-manuscript directory. All scripts below must be executed from this directory.

Pre-process the data.

Rscript data_analysis/preprocess.R

Run the analyses

# tower PCM
qsub data_analysis/run_r_script.sh data_analysis/split_pcm_da_manu.R

# tower GCM
qsub data_analysis/run_r_script.sh data_analysis/gcm_da_manu.R

# HRT
qsub data_analysis/run_r_script.sh data_analysis/hrt_da_manu_035.R

# PCM
qsub data_analysis/run_r_script.sh data_analysis/oat_pcm_da_manu_035.R

Make the results table

Rscript data_analysis/construct_results_table.R

Acknowledgments

We thank Timothy Barry and Ziang Niu for sharing their code and providing inspiration for our code pipeline.