Structural Modeling by using Overlapped Group Penalty

Fits a linear non-penalized phenotype (demographic) variables and penalized groups of prognostic effect and predictive effect, by satisfying such hierarchy structures that if a predictive effect exists, its prognostic effect must also exist. This package can deal with continuous, binomial or multinomial, and survival response variables, underlying the assumption of Gaussian, binomial (multinomial), and Cox proportional hazard models, respectively. It is implemented by combining the iterative shrinkage-thresholding algorithm and the alternating direction method of multipliers algorithms. The main method is built in C++, and the complementary methods are written in R.

smog

Structural Modeling by using Overlapped Group Penalty

Installation

# Install smog from CRAN
install.packages("smog")

# or install the source type package from GitHub:
# install.packages("devtools")
devtools::install_github("chongma8903/smog")

Features

• fits a linear non-penalized phenotype (demographic) variables and penalized groups of prognostic effect and predictive effect.
• satisfies such hierarchy structures that if a predictive effect exists, its prognostic effect must also exist.
• can deal with continuous, binomial or multinomial, and survival response variables.
• incorporates the iterative shrinkage-thresholding algorithm (ISTA) and the alternating direction method of multipliers algorithms (ADMM).

Usage

Create a new S3 class of smog, and the kernal function smog.default (or smog.formula) returns an object of the S3 class smog. The kernel functions include:

• smog.default: input the data and parameters to yield a model of the class smog.
• smog.formula: can accept formula to fit the model for the data.
• predict.smog: produces the predicted response values for new data, provided a fitted model.
• cv.smog: provides cross-validation analysis based on the data.
• cv.cglasso: cross-validation for conditional group lasso approach

Examples

sim = sim_rct_biomarker(n = 100, p = 20, p_prog = 2, p_pred = 2, p_both = 2)
y = sim$Y
x = sim$M
d = 20
g = c(d+1, rep(1:d,2))
v = c(rep(0,1), rep(1,2*d))
label = c("trt", rep(c("prog","pred"), c(d,d)))

sfit1 = cv.smog(x,y,g,v,label,family = "gaussian", type = "AIC")
plot(sfit1)

sfit2 = cv.cglasso(x,y,g,v,label,family = "gaussian", nlambda.max = 20)
plot(sfit2)