(Goeman, J.J., 2010. L(1) penalized estimation in the cox proportional hazards model. Biometrical Journal. Biometrische Zeitschrift, 52(1), 70-84.; R Development Core Team, 2009. R: A Language and Environment for Statistical Computing, Vienna, Austria. Available at: http://www.R-project.org.)
The Survival (SURV) module contains two functions for the analysis of censored survival data. The first is a basic comparison of the survival curves of two groups of samples. Sample data loaded into MeV is compared using the R package survival and the degree to which the curves differ is reported, along with a p-value.
The second feature of the module is the creation of a cox proportional hazards model based on the loaded gene expression data, using survival time as the reporting value. The R packaged penalized is used to build the model and to select the most informative genes. The L1-norm of the expression vector is used as a shrinkage parameter; users can select the lambda value. The module performs cross-validation of the model after it is built and reports a log likelihood as an indicator of model utility.
Censored survival data
No matter which of the analysis options you choose, you will need to select the sample annotation types that hold the survival data. There are two fields: survival time should point to a set of sample annotation that contains a floating-point number representing a time to event. The Censored field should point to a sample annotation field containing flags that indicate whether a given datapoint is censored or not. MeV can recognize several censoring flag types, including "Yes" or "No", "Censored/Uncensored", or 1/0. Values of "Yes", "Censored" or "1" indicate that the datapoint should be treated as a censored event, whereas "No", "Uncensored" and "0" are treated as valid, observed events.
After opening the SURV initialization dialog, select the type of analysis you want to run. Your options are:
After choosing the analysis type, click the Continue button to generate a cluster selection panel. This panel is different depending on which analysis type you have chosen.
A sample cluster selection panel will be generated. Select which samples belong in which groups and click ok to compare the survival profiles.
Use the cluster selection panel if you have previously made clusters and wish to run your analysis based on those clusters.
Use the drop-down boxes to choose which clusters’ samples are to be assigned to which time-points. For clusters you are not using, leave “unassigned”. The same condition and time-point requirements exist for this method, but the cluster selector makes for easier and more organized sample assignment.
A gene cluster selection panel with be generated. You can choose to select a group of genes to build the model with (faster) or choose to build the model based on all of the genes in the currently-loaded dataset (slow).
The Cox model will be build using an L1 Norm (lasso) penalty. Select the value of lambda to be used for this penalty. Higher values will result in faster-calculated models with fewer coefficients.
The SURV module outputs standard viewers and tables for MeV’s statistics analyses. It also creates survival curves for the selected sample groups.