README.md

Introduction

This repository contains the R software tools to run the PaRtitiOned empirical Bayes Ecm (PROBE) algorithm. We give a brief explanation of the method below. For more details see McLain, Zgodic, and Bondell (2022).

Minimal prior assumptions on the parameters are used through the use of plug-in empirical Bayes estimates of hyperparameters. Efficient maximum (MAP) estimation is completed through a Parameter-Expanded Expectation-Conditional-Maximization (PX-ECM) algorithm. The PX-ECM results in a robust computationally efficient coordinate-wise optimization, which adjusts for the impact of other predictor variables. The completion of the E-step uses an approach motivated by the popular two-groups approach to multiple testing. The PROBE algorithm is applied to sparse high-dimensional linear regression, which can be completed using one-at-a-time or all-at-once type optimization. PROBE is a novel alternative to Markov chain Monte Carlo (Liang et al. 2008; Bondell and Reich 2012 ; Chae, Lin, and Dunson 2019), empirical Bayes (George and Foster 2000; Martin, Mess, and Walker 2017; Martin and Tang 2020), and Variational Bayes (Carbonetto and Stephens 2012; Blei, Kucukelbir, and McAuliffe 2017; Ray and Szabó 2021) approaches to fitting sparse linear models.

Our proposed method performs Bayesian variable selection with an uninformative spike-and-slab prior on the regression parameters, which has not yet been used in the high-dimensional setting. We focus on MAP estimation of regression parameters, latent variable selection indicators, and the residual variance. We use a quasi Parameter-Expanded Expectation-Conditional-Maximization (PX-ECM) which is a combination of the ECM Dyk, Meng, and Rubin (1995) and PX-EM (Liu, Rubin, and Wu 1998). With the standard EM, the M-step would require optimization with respect to the high-dimensional regression parameter, which is not feasible with uninformative priors. Here, the ECM is used to break the M-step into a coordinate-wise optimization procedure. The PX portion of the algorithm adds a parameter which scales the impact of the remaining predictor variables. Unlike other coordinate-wise optimization approaches, this does not assume the impact of other predictor variables is fixed. The benefits of this formulation are that the impact of the remaining predictor variables only needs to be known up to a multiplicative constant, we can account for the increase in variability due to the dependence between the predictor and the impact of the remaining predictor variables, and it adds stability to the algorithm. The E-step consists of estimating the probability of the latent variable selection indicators via updating hyperparameters with plug-in empirical Bayes estimates, which is motivated by the popular two-groups approach to multiple testing (Efron et al. 2001; Sun and Cai 2007).

One-at-a-time and all-at-once variants of the PROBE algorithm have been developed. The examples below focus on the all-at-once variant.

Example of PROBE

The following is demonstration of how to implement all-at-once PROBE with simulated data. Please note that the package is still a work in progress and we are continually making improvements.

Install the package.

library(devtools)
install_github(repo="alexmclain/PROBE", subdir="probe")

Load the package and the data.

library(probe)
data(Sim_data)
attach(Sim_data) 
M <- dim(X)[2] 
M1 <- length(signal) 

Here, Sim_data contains the following elements:

• Y: vector of outcome variables for the training set,
• X: $n \times M$ covariate matrix for the training set,
• beta_tr: true value of beta coefficients, and
• signal: indicies of the $M_1$ non-null predictors

where $n=400$, $M=10000$ and $M_1=100$.

In this first run of the analysis we include the true signal ($\eta_i= X_i \beta$) and indices of the non-null beta coefficients (signal). This information will be used to create a plot of the MSE and the number of rejections with an indication of whether FDR$\leq \alpha$ by iteration ($t$).

eta_i <- apply(t(X)*beta_tr,2,sum) 
full_res <- probe( Y = Y, X = X, plot_ind = TRUE, 
                   eta_i = eta_i, signal = signal)

An example of performing prediction (of training data).

pred_res <- predict_probe_func( full_res, X = X, alpha = alpha)
head(pred_res)

How prediction intervals are constructed is discussed in Zgodic et al. (2023).

Re-run with test data

In this second run we’ll include test data instead of $\eta_i$ and signal. This run will give identical results to the first, but the plot will show the test MSE and the total number of rejections by iteration.

Load in some test data.

data(Sim_data_test)
attach(Sim_data_test) 
dim(X_test)
length(Y_test)

Here, Sim_data_test contains the following elements:

• Y_test: outcome variable for the test set, and
• X_test: covariate matrix for the test set.

Run using the test data. This information will be used to create a plot of the test MSE by iteration. This is for demonstration only. The test data does not select the number of iterations (the algorithm stops once the convergence criteria is met).

full_res <- probe( Y = Y, X = X, plot_ind = TRUE, 
                   Y_test = Y_test, X_test = X_test)

Predict for the test observations

pred_res_test <- predict_probe_func(full_res, X = X_test, 
                                       alpha = alpha)

## The true signal for test data
eta_test <- apply(t(X_test)*beta_tr,2,sum) 
plot(pred_res_test$Pred, eta_test)

Proportion of test PIs that contain the true signal.

(ecp_PI <- mean(1*I(Y_test>pred_res_test$PI_L & Y_test<pred_res_test$PI_U)))
detach(Sim_data)

See Zgodic et al. (2023) for a thorough examination of the coverage probabilities of the prediction intervals.

Example of PROBE with covariate data

This example will again fit a high-dimensional linear regression model, but this time additional covariates that are not subjected to the “sparsity” assumption are included. That is, we fit the model

$$Y_i = \mu + X_i \beta + Z_i \beta_Z + \epsilon$$

where $X$ has dimension $M \gg n$ and $Z$ has dimension $p < n$ (here $p=2$). How non-sparse covariates are brought into the model is discussed in Zgodic et al. (2023).

Load in the data:

data(Sim_data_cov)
attach(Sim_data_cov) 
M <- dim(X)[2] 
M1 <- length(signal) 
eta_i <- apply(t(X)*beta_tr,2,sum) 

Here, Sim_data contains the following elements:

• Y: vector of outcome variables for the training set,
• X: $n \times M$ predictor matrix for the training set,
• Z: $n \times 2$ covariate matrix for the training set (not subjected to the “sparsity” assumption),
• beta_tr: true value of $\beta$,
• beta_Z_tr: true value of $\beta_Z$,
• signal: indicies of the $M_1$ non-null predictors,

where $n=400$, $M=10000$ and $M_1=100$. The true signal term, $\eta_i$, only includes the impact of X (not of Z).

In this analysis we include $\eta_i$ and signal for plotting purposes.

full_res <- probe(Y = Y, X = X, Z = Z, plot_ind = TRUE, 
                  eta_i = eta_i, signal = signal)

Total number of iterations

full_res$count

Final estimates of the impact of Z versus the true values:

data.frame(true_values = beta_Z_tr, full_res$Calb_mod$res_data[-2,])

Compare to a standard linear model of $Z$ on $Y$:

summary(lm(Y~Z$Cont_cov + Z$Binary_cov))$coefficients

References

Blei, David M, Alp Kucukelbir, and Jon D McAuliffe. 2017. “Variational Inference: A Review for Statisticians.” Journal of the American Statistical Association 112 (518): 859–77.

Bondell, Howard D, and Brian J Reich. 2012. “Consistent High-Dimensional Bayesian Variable Selection via Penalized Credible Regions.” Journal of the American Statistical Association 107 (500): 1610–24.

Carbonetto, Peter, and Matthew Stephens. 2012. “Scalable Variational Inference for Bayesian Variable Selection in Regression, and Its Accuracy in Genetic Association Studies.” Bayesian Analysis 7 (1): 73–108. https://doi.org/10.1214/12-BA703.

Chae, Minwoo, Lizhen Lin, and David B Dunson. 2019. “Bayesian Sparse Linear Regression with Unknown Symmetric Error.” Information and Inference: A Journal of the IMA 8 (3): 621–53.

Dyk, David A. van, Xiao-Li Meng, and Donald B. Rubin. 1995. “Maximum Likelihood Estimation via the ECM Algorithm: Computing the Asymptotic Variance.” Statist. Sinica 5 (1): 55–75.

Efron, Bradley, Robert Tibshirani, John D. Storey, and Virginia Tusher. 2001. “Empirical Bayes Analysis of a Microarray Experiment.” J. Amer. Statist. Assoc. 96 (456): 1151–60. https://doi.org/10.1198/016214501753382129.

George, Edward I, and Dean P Foster. 2000. “Calibration and Empirical Bayes Variable Selection.” Biometrika 87 (4): 731–47.

Liang, Feng, Rui Paulo, German Molina, Merlise A Clyde, and Jim O Berger. 2008. “Mixtures of g Priors for Bayesian Variable Selection.” Journal of the American Statistical Association 103 (481): 410–23.

Liu, Chuanhai, Donald B. Rubin, and Ying Nian Wu. 1998. “Parameter expansion to accelerate EM: The PX-EM algorithm.” Biometrika 85 (4): 755–70. https://doi.org/10.1093/biomet/85.4.755.

Martin, Ryan, Raymond Mess, and Stephen G Walker. 2017. “Empirical Bayes Posterior Concentration in Sparse High-Dimensional Linear Models.” Bernoulli 23 (3): 1822–47.

Martin, Ryan, and Yiqi Tang. 2020. “Empirical Priors for Prediction in Sparse High-Dimensional Linear Regression.” J. Mach. Learn. Res. 21: 1–30.

McLain, Alexander C., Anja Zgodic, and Howard Bondell. 2022. “Sparse High-Dimensional Linear Regression with a Partitioned Empirical Bayes ECM Algorithm.” https://arxiv.org/abs/2209.08139.

Meng, Xiao-Li, and Donald B. Rubin. 1992. “Recent Extensions to the EM Algorithm (with Discussion).” In Bayesian Statistics 4, edited by J. M. Bernardo, J. O. Berger, A. P. Dawid, and A. F. M. Smith, 307–20. Oxford University Press.

———. 1993. “Maximum likelihood estimation via the ECM algorithm: A general framework.” Biometrika 80 (2): 267–78. https://doi.org/10.1093/biomet/80.2.267.

Ray, Kolyan, and Botond Szabó. 2021. “Variational Bayes for High-Dimensional Linear Regression with Sparse Priors.” Journal of the American Statistical Association 0 (0): 1–12. https://doi.org/10.1080/01621459.2020.1847121.

Sun, Wenguang, and T. Tony Cai. 2007. “Oracle and Adaptive Compound Decision Rules for False Discovery Rate Control.” J. Amer. Statist. Assoc. 102 (479): 901–12. https://doi.org/10.1198/016214507000000545.

Zgodic, Anja, Ray Bai, Jiajia Zhang, Yuan Wang, Chris Rorden, and Alexander McLain. 2023. “Heteroscedastic Sparse High-Dimensional Linear Regression with a Partitioned Empirical Bayes ECM Algorithm.” https://arxiv.org/abs/2309.08783.