Bioinformatics Stack Exchange is a question and answer site for researchers, developers, students, teachers, and end users interested in bioinformatics. It only takes a minute to sign up.
Sign up to join this community
What kind of multiple hypothesis testing correction does method="global" do in Limma's decideTests function? According to the documentation:
method="global" will treat the entire matrix of t-statistics as a single vector of unrelated tests.
and
method="global" is useful when it is important that the same t-statistic cutoff should correspond to statistical significance for all the contrasts.
What does this really mean? Does it use a Bonferroni correction?
More context from the docs, looking at the arguments to decideTests():
method: character string specifying how genes and contrasts are to be combined in the multiple testing scheme. Choices are "separate", "global", "hierarchical" or "nestedF".
adjust.method: character string specifying p-value adjustment method. Possible values are "none", "BH", "fdr" (equivalent to "BH"), "BY" and "holm". See p.adjust for details.
These are two different options.
One option (adjust.method) chooses the multiple test correction, for which the default is BH/FDR.
The other option (method) decides how the multiple test correction is applied, for example can it be applied separately to different genes or across contrasts. method = global in this context just means that every single test statistic is considered at the same time. An alternative to this would be to consider a nested design.
More intuitively, method = separate applies the correction to each column of your matrix independently, such that in a $m x n$ matrix of statistics, the "number of tests" is $m*n$ for global and $m$ for separate.
For understanding the effect of the number of tests on FDR, I suggest consulting the FDR wiki page. Basically, using a method other than global will increase power.
m and n correspond to here?
$\endgroup$
Aug 14, 2021 at 17:23
m would be number of rows of the test statistic matrix, n number of columns. So the total number of test stats in a single column is m, the total number of test stats in the full matrix is $m * n$. Columns of matrix correspond to slices of the analysis as defined by limma, IIRC. So you can choose to correct every possible test at once, or you can correct by slice of the analysis.
$\endgroup$
Aug 14, 2021 at 18:59