Rotation-based multiple testing in the multivariate linear model

Biometrics. 2014 Dec;70(4):954-61. doi: 10.1111/biom.12238. Epub 2014 Sep 30.

Abstract

In observational microarray studies, issues of confounding invariably arise. One approach to account for measured confounders is to include them as covariates in a multivariate linear model. For this model, however, the application of permutation-based multiple testing procedures is problematic because exchangeability of responses, in general, does not hold. Nevertheless, it is possible to achieve rotatability of transformed responses at the cost of a distributional assumption. We argue that rotation-based multiple testing, by allowing for adjustments for confounding, represents an important extension of permutation-based multiple testing procedures. The proposed methodology is illustrated with a microarray observational study on breast cancer tumors. Software to perform the procedure described in this article is available in the flip R package.

Keywords: Exchangeability; Familywise error rate; Microarray; Multiple testing; Permutation test; Rotation test.

MeSH terms

  • Algorithms*
  • Computer Simulation
  • Data Interpretation, Statistical
  • Gene Expression Profiling / methods*
  • Linear Models*
  • Multivariate Analysis*
  • Oligonucleotide Array Sequence Analysis / methods*
  • Reproducibility of Results
  • Rotation
  • Sensitivity and Specificity