[The problem of multiple testing and solutions for genome-wide studies]

Orv Hetil. 2005 Mar 20;146(12):559-63.
[Article in Hu]


The problem of multiple testing and its solutions for genome-wide studies. Even if there is no real change, the traditional p = 0.05 can cause 5% of the investigated tests being reported significant. Multiple testing corrections have been developed to solve this problem. Here the authors describe the one-step (Bonferroni), multi-step (step-down and step-up) and graphical methods. However, sometimes a correction for multiple testing creates more problems, than it solves: the universal null hypothesis is of little interest, the exact number of investigations to be adjusted for can not determined and the probability of type II error increases. For these reasons the authors suggest not to perform multiple testing corrections routinely. The calculation of the false discovery rate is a new method for genome-wide studies. Here the p value is substituted by the q value, which also shows the level of significance. The q value belonging to a measurement is the proportion of false positive measurements when we accept it as significant. The authors propose using the q value instead of the p value in genome-wide studies.

Publication types

  • English Abstract
  • Review

MeSH terms

  • Confidence Intervals
  • Data Interpretation, Statistical*
  • False Positive Reactions
  • Genomics*
  • Humans
  • In Situ Hybridization
  • Models, Statistical*
  • Research Design / standards*