Multiplicity, model validation, and reproducibility in high-dimensional microarray data

Project Heads:	Staff:	Students:
Vladimir Spokoiny	Elmar Diederichs	Le-Minh Ho

Phone: ++49 (0)30-20372 572,   Fax: ++49 (0)30-2044975

If you are looking for code packages, please go to the software area.

The main focus of the project is on the theory of multiple decisions and its applications to the analysis of medical data.

Recently, the rise of new research fields such as proteomics or metabolomics calls for new mathematical technologies to identify expression patterns or biomarkers. However, problems considered within this project are complicated due to the following reasons.

• Currently, multiplicity in statistical testing is still not fully understood and furthermore a challenge w.r.t. the reliability of the results.
• The high number of the genetic variants, probably associated with a desease, leads to high dimensional decision problems that are well known to be hardly analyzable due to computational as well as statistical limitations.

The new approach to deal with these difficulties favored by our research group is to translate the resulting decision problem into a model selection problem based on a detailed analysis of the risk function associated with this problem.