|
Abstract:
|
We discuss multiple comparison procedures adapted to the analysis of large dimensional phenotypes, as those collected in gene expression, proteomics, or metabolic studies. We underscore the importance of controlling error rates relative to the discovery of interest, which often are distinct from the rejection of single hypotheses. We illustrate the interest of a selective viewpoint, introduce novel error rate measure and controlling procedures and apply them to two datasets. These are results from long standing collaborations with many colleagues: it is meaningful to single out Y. Benjamini, M. Bogdan, M Bogomolov, E. Candes and C. Peterson.
|
