This is a collection of volcano plots with uncorrected or unspecified p-values that I have encountered in the wild.
Uncorrected hits can be found anywhere: lab meeting presentations, conference talks or posters, submitted manuscripts, or published papers.
How and why do uncorrected p-values contribute to the propagation of false knowledge?
Introduction
When testing many hypotheses, such as testing whether many proteins or RNA transcripts differ between two conditions, p-value correction makes it more likely that most of the statistically significant results are true. Sometimes, a researcher will not correct their p-values, which can contribute to propagation of false truths. Part of the problem is that there hasn’t been a thorough study of the sensitivity costs and false discovery rate (FDR) benefits of p-value correction, presented in a way that is intuitive to non-statisticians.
The SIMPLYCORRECT Web Application
To address this, with support and mentorship from Prof. Windy McNerney (Stanford U./Veterans Affairs), I made a web app that makes it easy to conceptualize and visualize the costs and benefits of p-value correction. The user puts in familiar experimental parameters such as the number of analytes, sample sizes, and variabilities, and a simulated omic experiment is performed. We call this the Simulator of P-Value Multiple Hypothesis Correction (SIMPLYCORRECT).
Click this button to go to the full-window app, or see the embedded app below.
The Publication
Using the 3 models developed for SIMPLYCORRECT, I conducted theoretical studies on the effects of various parameters on FDR and sensitivity in different contexts. A manuscript describing the conclusions of these studies can now be found in Analytical Chemistry. To access the article, click the link below.