Statistics / False positives and the q-value.

False-positives-and-the-qvalue.pdf

False Positives and the q-value에 대해 잘 설명해 놓은 문서.

A false positive finding in an experiment occurs when you conclude that something has an
effect on a variable when in fact it does not. There is a risk of this in all experiments, but
they can be a particular problem when tests are done on lots of variables (100's or 1000's)
such as happens in proteomics, transcriptomics or metabolomics. False discovery rates
(FDRs) and q-values are a way of quantifying the problem.
To see how they arise, we first consider an experiment in which there is no treatment
effect on any of the variables measured. In this case, any positive findings must be false.
For each of the many variables, we will have calculated a p-value. What will these pvalues
look like? It will be something like the plot below: they will be evenly spread
between 0 and 1. 5% of them will be less than p=0.05. These are the 5% false positives
present in every experiment, and they cannot be reduced by a bigger or better experiment
or a more careful analysis. ...