# TCGA EXAMPLE library(SSBoosting) # Load the data data(tcga.data) dim(tcga.data) # 578 observations and 8867 variables colnames(tcga.data)[1:25] # survival time, event indicator, demographic variables, study cohort indicator, and genes # Selecting stratification variable # Tumor size split on median size: tcga.data$size.ind <- 0 med <- median(tcga.data$Size) # same as if using full data tcga.data$size.ind[which(tcga.data$Size > med)] <- 1 strata.boosting(tcga.data$size.ind, tcga.data$survival_time) # Cancer stage strata.boosting(tcga.data$Stage, tcga.data$survival_time) # stage split into 9 groups # Combination of cancer stage and tumor size tcga.data$strata.ind <- 1 tcga.data$strata.ind[which(tcga.data$size.ind==0 & tcga.data$Stage <= 1)] <- 0 strata.boosting(tcga.data$strata.ind, tcga.data$survival_time) dim(tcga.data) # sample size is 578 # Variable selection covariates1 <- colnames(tcga.data) # 8859 genes + 5 clinical variables cov_test1 <- covariates1[-c(1,2,3, 8865, 8866)] # select possible covariates temp <- paste(cov_test1, collapse="+") temp2 <- paste("strata(strata.ind) +", temp,collapse = "+" ) # include stratification in formula fmla_test1_stratified <- formula(paste("Surv(survival_time, survival_ind) ~ ", temp2)) fit1 <- boosting_core(formula, data=tcga.data, rate=0.01, control=578) summary.boosting(fit1) plot.boosting(fit1, type="coefficients") # plot coefficient paths by boosting iteration plot.boosting(fit1) # plot of proportion selected by iteration