Simulations Comparing Left-Censored Imputation Methods

What is this?

This repository contains simulation scripts and artifacts accompanying Barbehenn and Zhao (2023) "Nonparametric empirical Bayes biomarker imputation and estimation" (https://arxiv.org/abs/2306.07239). The following methods are compared:

• EBM-Tobit [^1]
• Tobit maximum likelihood estimator [^1]
• GSimp [^2]
• QRILC [^3]
• zCompositions [^4]
• trKNN [^5]
• Empirical Bayes g-model with generalized exemplar support [^1]
• Empirical Bayes g-model with oracle support [^1]
• Vectorized empirical Bayes g-model with oracle support [^1]

The last two methods cannot be used in practice but are used to motivate the empirical Bayes framework for this imputation problem.

How to Set-up

Follow the following steps to set up the simulation environment. We used R version 4.2.2 (2022-10-31) in our simulations.

1. Set up the GSimp method
   • Clone the https://github.com/WandeRum/GSimp repository
   • Update line 3 in this repository's Utils.R to contain the location of the GSimp repository
   • Remove the following lines: lines 9-10 in GSimp.R, lines 7-8 in GSimp_evaluation.R, and lines 7-8 in Impute_wrapper.R
2. Ensure that the following packages are installed (later versions may work, but the following versions were used in the published simulations); most of the packages are on CRAN, but some are on BioConductor [1] ropls_1.30.0 reshape2_1.4.4 vegan_2.6-4 [4] lattice_0.21-8 permute_0.9-7 doParallel_1.0.17 [7] iterators_1.0.14 foreach_1.5.2 imputeLCMD_2.1 [10] impute_1.72.3 pcaMethods_1.90.0 Biobase_2.58.0 [13] BiocGenerics_0.44.0 norm_1.0-11.0 tmvtnorm_1.5 [16] gmm_1.7 sandwich_3.0-2 mvtnorm_1.1-3 [19] abind_1.4-5 missForest_1.5 magrittr_2.0.3 [22] FNN_1.1.3.2 rpart_4.1.19 glmnet_4.1-7 [25] Matrix_1.5-4.1 randomForest_4.7-1.1 ggplot2_3.4.2 [28] data.table_1.14.8 zCompositions_1.4.0-1 truncnorm_1.0-9 [31] NADA_1.6-1.1 survival_3.5-5 MASS_7.3-60 [34] ebTobit_1.0.1

How to Run Simulations

After setting up all of the simulations, use Rscript Run_Simulations.R to run the simulations. The number of simulations and combinations of censoring levels are defined on lines 122-128 of Run_Simulations.R. An additional log file containing environmental information is produced before simulations start.

Files

• EMBTobit.R: an R function implementing Algorithm 1 (the EBM-Tobit estimator in https://arxiv.org/abs/2306.07239)
• Utils.R: a helper R script to load the GSimp method and generate data for simulations
• Run_Simulations.R: an R script to reproducibly set up and run the simulations; both sims.csv and sessionInfo.log are produced by this script
• sims.csv: simulation results used for method comparison
• sessionInfo.log: a text file containing the full output of sessionInfo() used in for the simulations

References

[^1]: Barbehenn, A., & Zhao, S. D. (2023). Nonparametric empirical Bayes biomarker imputation and estimation. [^2]: Wei R, Wang J, Jia E, Chen T, Ni Y, et al. (2018) GSimp: A Gibbs sampler based left-censored missing value imputation approach for metabolomics studies. PLOS Computational Biology 14(1): e1005973. https://doi.org/10.1371/journal.pcbi.1005973 [^3]: Lazar, C., & Burger, T. (2022). imputeLCMD: A Collection of Methods for Left-Censored Missing Data Imputation. https://CRAN.R-project.org/package=imputeLCMD [^4]: Palarea-Albaladejo, J., & Martín-Fernández, J. A. (2015). zCompositions — R package for multivariate imputation of left-censored data under a compositional approach. Chemometrics and Intelligent Laboratory Systems, 143, 85–96. https://doi.org/https://doi.org/10.1016/j.chemolab.2015.02.019 [^5]: Shah, J. S., Rai, S. N., DeFilippis, A. P., Hill, B. G., Bhatnagar, A., & Brock, G. N. (2017). Distribution based nearest neighbor imputation for truncated high dimensional data with applications to pre-clinical and clinical metabolomics studies. BMC Bioinformatics, 18(1), 114. https://doi.org/10.1186/s12859-017-1547-6