Significance testing using LASSO

Using cross validation

The current strategy we use to select gene sets that strongly predict the response variable is cross validation. Mean squared error is determined across values of lambda. non-zero betas at a lambda one standard error above the minimum are chosen as significant predictors.

Using covariance test statistics

covariance test statistics for LASSO implemented in covTest allows p-values to be determined for individual predictors.

covTest currently struggles at the scale of data used for GSEAMA, but on simulated data, variables selected by either CV or covTest were similar.

a_covTest <- covTest::covTest(lars::lars(as.matrix(m@matrix), y, type = "lasso"), as.matrix(m@matrix), y)
m@colData$p.value = a_covTest$results[,3]