SmulTCan

The SmulTCan app allows users to analyze groups of genes, miRNAs or gene-level CNVs or methylation levels for which they have downloaded values and sample information from UCSC Xena (http://xena.ucsc.edu), with regard to the four types of survival data; overall, disease-specific, disease-free, and progress-free on TCGA (https://www.cancer.gov/tcga). The TSV file obtained from Xena should be in the following structure: “sample” and “samples” for the first two columns, followed by expression columns for the genes of interest, and “cancer type abbreviation” and “sample_type” for the last two columns. Xena uses \(\log_2(norm\_value+1)\) gene and miRNA expressions (a normalized form). Ideally, the TSV file containing either gene or miRNA expressions or gene-level CNVs or methylation \(\beta\)-values should be generated through selecting TCGA-PANCAN as the study type in Xena. When using gene-level methylation \(\beta\)-values, values in the downloaded TSV file should be multiplied by 100 before upload onto the app by the user for better interpretation of results. Only primary (or primary blood derived for LAML) tumor samples and their corresponding survival information are used in the analyses in the app for each TCGA dataset.

The app uses the Cox proportional hazards (CPH) model to analyze the impacts on survival of input genes multivariably. Once the input TSV file with the required Xena format is uploaded, the user can select/deselect names of the genes with expressions inside the input file from the side-bar panel. Several best subset selection methods are incorporated in the app, which can be made directly through the CPH model by picking genes whose coefficients have \(p < 0.05\), or by applying additional algorithms such as “glmnet” or “BeSS”. The “glmnet” tab has options for elastic net regularization, where lasso (= 1) is the default method and users can drag the slider all the way through to ridge (= 0) regression. By default, “glmnet” uses 10-fold cross-validation, however users can also choose the sample size of each dataset for the number of folds. Also included in the app is the “BeSS” package for best subset selection using ridge regression. The K-M plots in the app allow users to view results of the prognostic risk analyses based on a median prognostic index (PI) calculated from the corresponding best subset selection method.