Shrinkage t Statistic and Correlation-Adjusted t Score

This package implements the "shrinkage t" statistic introduced in Opgen-Rhein and Strimmer (2007) and a shrinkage estimate of the "correlation-adjusted t-score" (cat score) described in Zuber and Strimmer (2009). It also offers a convenient interface to a number of other regularized t-statistics commonly employed in high-dimensional case-control studies.

Version: 1.1.4

Authors: Rainer Opgen-Rhein, Verena Zuber, and Korbinian Strimmer.

Documentation and Installation:

• Help pages and release history.
• Local download of st source package.
• Licensed under the GNU GPL version 3 (or any later version).

Additional Information:

• The "shrinkage t" statistic compared to a number of other regularized t-type statistics, in terms of quality of ranking and ROC curves (from Opgen-Rhein and Strimmer 2007): (click on image to enlarge):
  
• SUMMARY:
  If variances across genes are similar (left column), fold-change-type statistics work best. Conversely, if variances are heterogeneous (right column), t-type statistics perform best.
• Regularized t-scores can automatically adapt to the variance scenario at hand. "shrinkage t", "SAM", and "moderated t" are the only three consistently top-performing test statistics.
• The "shrinkage cat score" ("correlation-adjusted t score") is a decorrelated version of the shrinkage t score, and thus estimates the contribution of each individual gene in separating the two groups, after removing the effect of all other genes.

Relevant Papers:

• Opgen-Rhein, R., and K. Strimmer. 2007. Accurate ranking of differentially expressed genes by a distribution-free shrinkage approach. Statist. Appl. Genet. Mol. Biol. 6: 9. (reprint)
• Zuber, V., and K. Strimmer. 2009. Gene ranking and biomarker discovery under correlation. Bioinformatics 25: 2700-2707. (reprint, arXiv:0902.0751)

Back to software page.