MR.ASH (github: mr-ash-workflow)

Numerical experiments for evaluating the "multiple regression with adaptive shrinkage" (MR.ASH) method and other linear regression methods that are well suited for large-scale data sets.

This workflow is for reproducing the results used in the manuscript

MR.ASH: A Novel Variational Empirical Bayes Approach to Multiple Linear Regression.

---

Vignette

See Mr.ASH Vignette

---

Reproducing the result using R

The proposed VEB (Variational Empirical Bayes) approach can be implemented via our companion package mr.ash.alpha, which can be found in MR.ASH github.

Quick Start for mr.ash.alpha

To install the latest version of the R package mr.ash.alpha from MR.ASH github, clone or download the git repository, then use the install_local function from devtools, replacing "path/to/repo" with the appropriate directory path:

devtools::install_local("path/to/repo")

Also, one can install the package via

devtools::install_github("stephenslab/mr.ash.alpha")

Quick Start for Other comparision methods

The list of competitors is as follows.

# Penalized linear regression based on Ridge, Lasso, Elastic Net penalties, fitted by CV (Ridge, Lass, E-NET)
install.packages("glmnet")
# Penalized linear regression based on SCAD, MCP penalties, fitted by CV (SCAD, MCP)
install.packages("ncvreg")
# Penalized linear regression based on L0 penalty (or L0+L1, L0+L2), fitted by CV (L0Learn)
install.packages("L0Learn")
# Bayesian linear regression based on additive single effect model, fitted by VB (SuSiE)
install_github("stephenslab/susieR")
# Bayesian linear regression based on hieararchical spike-and-slab prior, fitted by MCMC (BayesB, BLasso)
install.packages("BGLR")
# Bayesian linear regression based on hieararchical spike-and-slab prior, fitted by VB and discrete BMA (varbvs)
install.packages("varbvs")

For detailed discussion on the comparison methods, please see the package documentations and reference therein.

Implementation of the methods

For each method, its implementation has a wrapper in code/method_wrapper.R.

For each simulation setting, its implementation has a wrapper in code/sim_wrapper.R.

Results

We provide *.Rmd files for reproducing all the simulations in analysis folder.

We provide *.RDS files for storing all the results from analysis/*.Rmd in results folder.

We provide analysis/plots_for_the_paper.RDS for reproducing all the figures in our manuscript MR.ASH.

Figures

Figure	Description	Source code for reproducing the result	Source code for plotting
Figure 1	Flexibility of MR.ASH prior, in terms of shrinkage operator and penalty	analysis/Figure1_Sourcecode	analysis/Figure1_Plot
Figure 2 and Figure 3	Adaptation to Sparsity (Penalized Linear Regression and Bayesian Linear Regression)	analysis/Figure2_Sourcecode1 analysis/Figure2_Sourcecode2	analysis/Figure2_Plot and analysis/Figure3_Plot
Figure 4	Sample size, signal strength, signal shape	analysis/Figure4_Sourcecode1 analysis/Figure4_Sourcecode2	analysis/Figure4_Plot
Figure 5	Computation time (scalability)	analysis/Figure5_Sourcecode	analysis/Figure5_Plot
Figure 6	Relative Performance (boxplot)	analysis/Figure6_Sourcecode	analysis/Figure6_Plot
Figure 7	Initialization and Update Order	analysis/Figure7_Sourcecode	analysis/Figure7_Plot

---

Reproducing the results using DSC

Not ready yet

---

License

Copyright (c) 2019-2020, Youngseok Kim.

All source code and software in this repository are made available under the terms of the MIT license. See file LICENSE for the full text of the license.

---

Credits

The mr.ash R package was developed by Youngseok Kim.