Discovery Screen

Discovery screen

The Discovery tab tests many candidate modules for coordinated change. Its purpose is to find groups of related genes whose members each change modestly, so that few pass a per-gene test, but which together show a consistent shift.

Candidate modules

Source	Definition
Gene families	Genes sharing a symbol stem (for example Skint1 to Skint11)
Loaded GMT library	Each set in the library
HGNC gene groups	Curated human gene groups; for non-human data, matched by case-normalized symbol (an orthology approximation)

Modules are tested if their size in the filtered data lies within the size limits (default 3 to 100).

Module score

For each sample, the module score is the mean z-score of the module's genes, with each gene standardized across the included samples.

Tests

Design	Test
Two groups	CAMERA (limma), a competitive test that inflates the variance by the estimated inter-gene correlation. Parametric; no permutation floor.
Three or more groups	Seeded permutation test against size-matched random modules. The inter-gene correlation is shown for each module as a caution, because random-set nulls understate it.

FDR is controlled by Benjamini–Hochberg across all candidates. Overlapping modules are positively dependent, a condition under which BH remains valid.

Result tiers

• Significant: FDR at or below 0.05.
• Hidden module: significant as a module while fewer than 25% of members are individually significant.
• Nominal tier: shown only when no module survives FDR; modules with nominal p at or below 0.01, labelled as nominal. These are ranked hypotheses, not findings.

Robustness

A leave-one-out note appears on a result card when the direction of the module's effect depends on a single sample.

Interpretation

Structural gene families (for example DEAD-box helicases) can move together because of global changes in cell composition or proliferation rather than a shared regulatory programme. Functionally coherent families are stronger candidates. Confirm results in an independent dataset before reporting them as findings.

Reference

Wu D and Smyth GK (2012) Nucleic Acids Res 40:e133.