Chapter 13 Multiple testing
When performing a large number of tests, the type I error is inflated: for α=0.05 and performing n tests, the probability of no false positive result is:
0.095 x 0.95 x … (n-times) <<< 0.095
The larger the number of tests performed, the higher the probability of a false rejection!
Many data analysis approaches in genomics rely on itemby-item (i.e. multiple) testing:
Microarray or RNA-Seq expression profiles of “normal” vs “perturbed” samples: gene-by-gene
ChIP-chip: locus-by-locus
RNAi and chemical compound screens
Genome-wide association studies: marker-by-marker
QTL analysis: marker-by-marker and trait-by-trait
False positive rate (FPR) - the proportion of false positives among all resulst.
False discovery rate (FDR) - the proportion of false positives among all significant results.
Example: 20,000 genes, 100 hits, 10 of them wrong.
FPR: 0.05%
FDR: 10%
13.1 The Bonferroni correction
The Bonferroni correction sets the significance cut-off at α/n.