Multiple Testing in Large-Scale Genetic Studies

Recent advances in Molecular Biology and improvements in microarray and sequencing technologies have led biologists toward high-throughput genomic studies. These studies aim at fi nding associations between genetic markers and a phenotype and involve conducting many statistical tests on these markers. Such a wide investigation of the genome not only renders genomic studies quite attractive but also lead to a major shortcoming. That is, among the markers detected as associated with the phenotype, a nonnegligible proportion is not in reality (false-positives) and also true associations can be missed (false-negatives). A main cause of these spurious associations is due to the multiple-testing problem, inherent to conducting numerous statistical tests. Several approaches exist to work around this issue. These multiple-testing adjustments aim at de fi ning new statistical con fi dence measures that are controlled to guarantee that the outcomes of the tests are pertinent.The most natural correction was introduced by Bonferroni and aims at controlling the family-wise error-rate (FWER) that is the probability of having at least one false-positive. Another approach is based on the false-discovery-rate (FDR) and considers the proportion of signi fi cant results that are expected to be false-positives. Finally, the local-FDR focuses on the actual probability for a marker of being associated or not with the phenotype. These strategies are widely used but one has to be careful about when and how to apply them. We propose in this chapter a discussion on the multiple-testing issue and on the main approaches to take it into account. We aim at providing a theoretical and intuitive de fi nition of these concepts along with practical advises to guide researchers in choosing the more appropriate multiple-testing procedure corresponding to the purposes of their studies.