Inflated false discovery rate due to volcano plots: problem and solutions.

Area of Science:

• Statistics
• Bioinformatics
• Genomics

Background:

• Volcano plots are popular for selecting significant findings after multiple testing procedures like Benjamini-Hochberg (BH).
• This selection method involves filtering for small adjusted P-values and large effect sizes.
• However, BH does not guarantee error control on these filtered subsets, potentially inflating false discoveries.

Purpose of the Study:

• To demonstrate the inflated Type I error rate associated with volcano plot selection.
• To introduce and validate alternative double filtering methods that maintain false discovery rate control.
• To provide a practical, accessible tool for reliable feature selection.

Main Methods:

• Simulation experiments to quantify Type I error inflation.
• Analysis of RNA-sequencing data to illustrate real-world impact.
• Development and evaluation of novel double filtering procedures for multiple testing.
• Implementation of the proposed methods in an interactive web application.

Main Results:

• Volcano plot selection substantially inflates the Type I error rate.
• Features with the largest estimated effect sizes are frequently false positives.
• The proposed alternative methods effectively control the false discovery rate.
• The interactive web application provides a user-friendly interface for the new procedures.

Conclusions:

• Standard volcano plot selection is unreliable due to inflated false discovery rates.
• The developed methods offer a statistically sound alternative for robust feature selection.
• The publicly available web application facilitates the adoption of these improved techniques in research.