signs   -  


SignS: molecular signatures with survival and censored data (v.20070801)

SignS is a web tool for gene selection and signature finding in problems where the dependent variable is patient survival or, more generally, a right-censored variable. Four approaches have been implemented: the threshold gradient descent method of Gui & Li (2005), a method that uses a combination of gene filtering, clustering and survival model building, very similar to the one used in Dave et al. (2004), a method that uses random forests with conditional inference tress by Hothorn et al. (2006a), and a method that uses boosting with component-wise univariate Cox models (Hothorn et al., 2006a).

To use this web tool you need to provide three files, one with the gene expression data, another with the survival time (time to event, e.g., time to death) and a a third one indicating whether the event was observed or not (the later are the censored cases).

If you have a validation data set, you can also include the expression data, the survival time and the censored status of the validation data. However, DO NOT (I repeat DO NOT) keep playing around with the parameters until things look nice with the validation data. That would be a serious case of overfitting.

Input files (help)

Expression data file (?)
Survival time file (?)
Survival status (event) file (?)

Validation data (?)

Use validation data

Validation covariate file:
Validation survival time file:
Validation survival status (event) file:

Method (help)

FCMS: Filter, Cluster, and Stepwise model selection (as in Dave et al.).

Minimum gene-wise Cox p-value in gene filtering step
Max. number of genes in a cluster
Min. number of genes in a cluster
Min. correlation of genes in a cluster

TGD: Threshold Gradient Descent (Li and Gui).

Maximum number of iterations
∆ν (Delta nu)
τ (tau) [Please, use only large values, over 0.8, unless you really know what you are doing!]

Random forests: Random forests using conditional inference trees (Hothorn et al., 2006a).

Number of genes

GLMboost: Boosting of component-wise Cox models (Hothorn et al., 2006 b).

Note: Floating-point numbers, such as 0.001, can be specified as either "0.001", "1e-3" or "1E-3" (without the "", of course). Anything else will produce an error.

Click "Submit" to start execution.


Citing SignS

If you use SignS, please cite it in your publications. Please provide the URL and the publication:

Diaz-Uriarte, R. 2008. SignS: a parallelized, open-source, freely available, web-based tool for gene selection and molecular signatures for survival and censored data. BMC Bioinformatics 2008, 9:30.

[Python Powered] [R Project for Statistical Computing]

Send comments to webmaster