Combining effect size and multiple-testing evidence to identify genes worth following up.
The volcano plot summarizes a differential expression analysis. The MA plot shows abundance vs. change; the volcano plot puts change next to statistical evidence, so a gene is judged on both effect size and reliability.
Two axes:
Most genes cluster at the base; interesting ones rise into the upper corners, hence the volcano shape.
A candidate usually has to pass two filters: a minimum fold change and a maximum adjusted p-value. The pair filters out both noisy large changes and tiny changes that become statistically detectable only because the experiment is precise.
Figure 1. Volcano plot of 800 simulated genes. Upper-right: up in treated and FDR-significant; upper-left: down in treated and FDR-significant.
High fold change without FDR support often means there is too much uncertainty. A very small adjusted p-value with tiny fold change may be statistically real but still not biologically meaningful for the question at hand.
Each filter alone is insufficient: fold change alone captures noisy outliers; FDR alone may include genes with negligible expression changes.
Figure 2. Three filtering strategies. The combined approach (right) isolates genes that satisfy both criteria.
Labels are most helpful when they are selective. Here the labeled genes are ranked among those that pass both filters, so the labels point to candidates rather than merely the smallest adjusted p-values.
Figure 3. Ranking by significance and effect size.