This directory contains the codes to perform all the analyses and reproduce all the figures presented in the paper: Wong KY, Fan C, Tanioka M, Parker JS, Nobel AB, Zeng D, Lin DY, Perou CM. I-Boost: an integrative boosting approach for predicting survival time with multiple genomics platforms. 2019.
The R-package IBoost can be found at https://github.com/alexwky/I-Boost.
Before running the codes, download the processed TCGA pan-cancer data set TCGA_8cancer_rmmis.csv from https://doi.org/10.5281/zenodo.2530387 and store it in the directory Data.
The simulation data sets used in the paper are stored in the directory SimulationData. The zip files in the directory contain the file allPredictorValues.csv and 3,000 files with names in the form of Data-setting[setting number]-[replication number].csv that consist of rows in the form:
0.240290380066643,0,276
0.578869428061267,1,531
2.14527656943497,1,837
0.272058507083588,1,1241
.
.
.
Each of the 3,000 files contains the simulated data for 500 subjects for a specific setting and simulation replicate. Each row in a file contains data for a subject. The first element on each row is the survival or censoring time, the second element is the event indicator that equals 1 or 0 if the event is observed or right-censored, respectively, and the third element is the index of the row in the file allPredictorValues.csv that contains the values of the clinical/genomic predictors of this subject. For instance, a value of 1 for the third element represents that the values of the clinical/genomic predictors of this subject are stored in the first row of allPredictorValues.csv.
Run the bash code simulations.sh in the home directory. It creates the directories SimulationData, SimulationResults, SimulationSettings, and R codes in the Programs directory. Run the program sim_setting.R in the Programs directory; it will generate files that contain parameter values for simulating data. Run the program gendata.R in the Programs directory; it will generate 1,000 simulated data sets for each simulation setting in the directory SimulationData (and the file allPredictorValues.csv described above). (This step is not necessary for the simulation studies, because the (same) data sets are generated internally in the simulation programs.)
Run the R programs sim-[method name]-s[setting].R in the Programs directory. Each program performs analysis on 1,000 simulated data sets and output the results in the directory SimulationResults. Each row of the output files represents:
replication number, number of variables selected, number of true signal variables selected, risk correlation, number of clinical variables selected, number of gene modules selected, number of protein expressions selected, number of miRNA expressions selected, number of mutations selected, number of copy number variations selected, MSE for clinical variables, MSE for gene modules, MSE for protein expressions, MSE for miRNA expressions, MSE for mutations, MSE for copy number variations
Some of the programs may take very long time to run. To save time, one may run the analysis for separate replicates in separate programs and combine the results in output files in the format given above.
Run the bash codes analysis_splits.sh, analysis_splits_cox.sh, and analysis_wholedata.sh in the home directory. They will create all programs necessary to perform all analyses on the TCGA data sets presented in the paper. Run the created programs in the Programs directory. The programs with prefix DataAnalysis- in the name will perform analyses over 30 training/testing data splits using maximum likelihood estimation, LASSO, elastic net, I-Boost-CV, or I-Boost-Permutation. The results will be written in corresponding directories in Results. Files with prefix Model_ in the name contain the selected predictors and the estimated regression parameters, and files with prefix summary_ in the name contain summaries of the analysis results, including the C-index in the testing set. Upon completion of all analyses, run the program gather.R in the Programs directory, which will combine the summary_ files.
The programs with prefix WholeDataAnalysis- in the name perform the analyses on the whole LUAD, KIRC, and pan-cancer data sets (without setting aside subjects for testing). The selected predictors and estimated regression parameters are written in corresponding directories in Results/WholeData.
Upon completion of all analyses, run the programs plotFigure1.R, plotFigure2and3andS1.R, plotFigure4.R, plotFigure5.R, and plotFigure6.R in Programs. They will generate Figures 1-6 and Figure S1 in the directory Plots.