Choose the analysis engine
The same biological question can be asked with different evidence models. Choose the engine before choosing a database: ORA tests overlap in a thresholded list, GSEA tests a coordinated shift across a ranking, and network-aware or weighted methods add information that the classical tests do not model.
Decision sequence
- Do you have a thresholded list or a complete ranking?
- Is the sign of the statistic meaningful, and do you need both directions?
- Is there one list or several groups/conditions?
- Do detection bias, network topology, or multiple omics layers carry important evidence?
- Which result object should be preserved for comparison, visualization, and interpretation?
Start with the enrichment terminology, then compare ORA and GSEA. The same result can later be routed to knowledge sources in Part 2 and to evidence integration in Part 3.
| Analysis question | Engine | Main chapter |
|---|---|---|
| Are selected genes over-represented? | ORA / hypergeometric test | ORA |
| Is a gene set shifted across the full ranking? | GSEA | GSEA |
| Do several clusters or conditions have different profiles? | compareCluster() |
Comparing biological themes |
| Does network topology or detection bias matter? | NSEA / weighted enrichment | Network and weighted methods |
| Do several molecular layers contribute evidence? | Multi-omics aggregation and fusion | Multi-omics section |
The statistical engine determines what is tested. The knowledge source determines what the tested terms mean; those two responsibilities are intentionally separated in the next part.