Skip to contents

Analysis types

Source: vignettes/ORA_vs_ES.Rmd
ORA_vs_ES.Rmd

ORA vs Enrichment (two ways to run TaxSEA)

TaxSEA supports two complementary analysis modes: Enrichment and ORA (Over-Representation Analysis).
They answer related but distinct questions and are appropriate in different situations.

Enrichment (rank-based)

What it does (in plain terms)

You provide TaxSEA with a ranked list of taxa (for example from most increased to most decreased).
TaxSEA then asks:

Do the members of a given taxon set tend to appear unusually high or low in this ranking?

Importantly, all taxa are used, not just those passing a significance threshold.

Use Enrichment when:

  • You have differential abundance results (e.g. log fold changes, test statistics, signed p-values)
  • Taxa can be meaningfully ordered by strength and direction of effect
  • You want to avoid hard cutoffs such as “significant vs not significant”

Pros

  • Uses all taxa, not only a selected subset
  • Less sensitive to arbitrary thresholds
  • More stable when signal is spread across many taxa
  • Can detect subtle, coordinated shifts across a taxon set

Cons

  • Requires a sensible ranking metric
  • Not appropriate if the data are strictly presence/absence with no meaningful ordering

ORA (Over-Representation Analysis)

What it does (in plain terms)

You provide TaxSEA with a list of taxa of interest (your “hits”), along with a background universe.
TaxSEA then asks:

Are taxa from this set appearing in my hit list more often than expected by chance?

This is a classic enrichment-of-hits approach using contingency tables.

When to use it

Use ORA when:

  • Your data are naturally binary (presence vs absence, detected vs not detected)
  • A ranked statistic is not meaningful or not available
  • You have a strong, biologically justified threshold defining your taxa of interest

Pros

  • Simple and intuitive interpretation
  • Works well for presence/absence style data
  • Appropriate when only a hit list is available

Cons

  • Highly sensitive to how “hits” are defined
  • Discards information about effect size and ordering
  • Results can change substantially with small changes to thresholds

In most microbiome differential abundance analyses, Enrichment is recommended whenever a reasonable ranking can be constructed (e.g. log fold change, Wald statistic, signed −log10 p-value). It makes fuller use of the data and avoids arbitrary cutoffs.

ORA should be used when ranking is not possible or not meaningful, particularly in presence/absence scenarios or when the scientific question explicitly concerns membership in a predefined hit list rather than graded shifts.

Both methods are valid; the choice depends on the structure of the data and the biological question being asked.