stabilityBIG: Computes the area under the stability path for all covariates

Description

This function implements the same model selection technique extensively described in stabilityGLM. The sole difference is the use of a different elastic net solver. In this function, we make use of biglasso. Thanks to its parallel backend, biglasso scales well to high-dimensional GWAS datasets. However, in our case, because of the use of additional backend files, a slight decrease in runtime is to be expected, compared with stabilityGLM.

Usage

stabilityBIG(X, Y, family = "gaussian", n_subsample = 20,
  n_lambda = 100, lambda_min_ratio = 0.01, eps = 1e-05,
  short = TRUE, ncores = 2)

Arguments

design matrix formatted as a big.matrix object

response vector

family

response type. Either 'gaussian' or 'binomial'

n_subsample

number of subsamples for stability selection

n_lambda

total number of lambda values

lambda_min_ratio

the minimum value of the regularization parameter lambda as a fraction of the maximum lambda, the first value for which the elastic net support is not empty.

eps

elastic net mixing parameter (see stabilityGLM for more details)

short

whether to compute the aucs only on the first half of the stability path. We observed better performance with thresholded paths

ncores

number of cores for the biglasso solver

Value

a vector grouping the aucs of all covariates within X

References

Slim, L., Chatelain, C., Azencott, C.-A., & Vert, J.-P. (2018). Novel Methods for Epistasis Detection in Genome-Wide Association Studies. BioRxiv.

Meinshausen, N., & B<U+00FC>hlmann, P. (2010). Stability selection. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 72(4), 417<U+2013>473.

Haury, A. C., Mordelet, F., Vera-Licona, P., & Vert, J. P. (2012). TIGRESS: Trustful Inference of Gene REgulation using Stability Selection. BMC Systems Biology, 6.

Examples

Run this code

# NOT RUN {
n <- 100
p <- 25
X <- bigmemory::as.big.matrix(matrix(runif(n * p), ncol = p))
Y <- runif(n, min = 0, max = 1) < 0.5
aucBIG <- stabilityBIG(X, Y,
  family = "binomial", short = TRUE,
  ncores = 1, n_lambda = 200, n_subsample = 1
)
# }