Learning center

Robinson–Foulds: topology differences, normalization and saved tables

Interpret unrooted split counts, the fixed normalization denominator and observed percentages, using retained small-tree comparisons and complete saved tables.

Robinson–Foulds distance counts differences in tree topology. Read the split definition, taxon set and normalization before comparing scores; a percentage displayed beside a distance is not a probability or a percentile.

This guide covers the Robinson-Foulds and Normalized RF routes in Tree Comparison & Topology Metrics. It reuses seven small pair contracts and one saved four-taxon native workflow. The app’s other routes, sequence-based tree inference and optional null-distribution calculations have separate evidence requirements.

1. Define the pair before comparing it

Keep each original Newick tree, its name and the input order. The supplied-tree path in the dated source snapshots reviewed below requires at least four taxa and matching taxon sets across the supplied trees. The pairwise RF routes use the first two trees. All twenty app routes start selected in this source version; explicitly retain the selected-method list for the calculation you intend.

InspectMeaning for these routes
Taxon identitiesUse consistent, unique leaf labels for the intended same taxa. Do not interpret a missing or renamed taxon as a topology change. The retained checks use only A, B, C and D; they do not establish malformed-label handling.
First and second treesPreserve their order and names. Some figures, logs and fields still say NJ and UPGMA: with supplied trees, those labels refer to the first and second inputs, not evidence that either tree was inferred by that method.
Topology and branch lengthsThese RF routes use unrooted nontrivial split sets. Branch-length changes alone do not change the split count. A branch-length comparison needs its own definition and guide.
Delivery and execution scopeThe saved native workflow receives its pair when the app opens, then exercises Run, viewer and Export All. It does not verify a native Load dialog, launcher, hosted upload or sequence-inference path.

The source accepts named Newick collections through its data contract. Before an authorized run, confirm that the available delivery path supplies that contract; this lesson does not provide evidence for an untested file-upload action. Preserve any refused or unavailable outcome with its reason.

2. Count the splits that distinguish the trees

Removing an internal edge separates the leaves into two groups. In this implementation a nontrivial split has at least two taxa on each side; the two sides are unordered. Each unique split is counted once. Raw RF is the number of splits present in exactly one of the two sets: the size of their symmetric difference.

For the retained four-taxon example, one tree has split AB | CD and the other has AC | BD. They share no nontrivial split; one split is unique to each tree, so the retained raw RF is 2. This is a count with no physical unit, not two substitutions or a measure of elapsed evolutionary time.

The DendroPy comparison reference also distinguishes unweighted symmetric difference from branch-weighted distances. That reference supplies terminology; it does not establish that Quoneon executes DendroPy or has equivalent behavior on every tree.

3. Keep the denominator with the score

This route uses normalized RF = raw RF / [2(n − 3)], where n is the number of taxa. The denominator is the binary unrooted-tree upper bound for that taxon count. It is not the sum of the observed split counts, nor a bound adjusted to the resolution of the particular pair.

With four taxa the denominator is 2. The discordant resolved pair gives 1.0000. A star tree with no nontrivial splits compared with the resolved tree gives raw RF 1 and normalized RF 0.5000. Two identical star trees give zero despite containing no resolved internal relationship. Zero distance establishes agreement of these split sets; it does not establish resolution, correctness or biological truth.

4. Read the retained examples without inflating coverage

The seven pair contracts were previously checked in both operand orders. Reversing a pair is a symmetry check, not another independent dataset. The following values are read from those saved receipts; no RF calculation was repeated for this guide.

Retained pairRaw RFNormalized RF
Same resolved tree twice00.0000
Discordant resolved trees21.0000
Root-child order swapped00.0000
Same unrooted topology, rerooted representation00.0000
Only terminal branch lengths changed00.0000
Star tree and resolved tree10.5000
Same star tree twice00.0000

The native workflow reuses the discordant pair: the selected original application produced raw RF 2 and normalized RF 1, then saved both tables and figures. The retained checks do not cover larger or real-data trees, arbitrary Newick syntax, missing taxa, duplicate labels or inferred trees. The source’s simple parser is not a general Newick-conformance certificate.

5. Separate observed distance from statistical evidence

The corrected view shows the observed normalized distance on a 0–1 axis. Its log labels Normalized distance percentage, which is the observed normalized RF multiplied by 100. Thus 100.0% for the discordant pair is not a 100th percentile, a p-value, confidence in a topology or evidence that either tree is correct.

No null distribution was supplied in the accepted example. The optional panel reports its absence and the saved P-value field is not_uploaded. Keep that text as missing optional input; do not turn it into zero or infer significance. Optional null-distribution, empirical-tail and interval behavior is outside this guide’s accepted workflow evidence.

An older retained version displayed an unsupported expected-distance curve and used the misleading log label “Percentile.” The accepted correction replaced that display with the observed distance and a truthful percentage label. The earlier failure remains historical evidence; it is not the state of the corrected method described here.

6. Read all saved fields and their precision

The accepted native Export All saved two figures, two CSV tables and two text artifacts, plus a separate integrity manifest: seven files. The two CSVs have a Parameter/Value header and five data rows each. Read both complete tables alongside the input-and-methods record.

ArtifactWhat to retain
Raw RF CSVRF Distance, Shared Splits, Max RF (2n−6), NJ Unique and the reference. NJ Unique means the first input’s unique-split count; the second input’s unique count is not a separate saved row in this table.
Normalized RF CSVNormalized RF, Raw RF, Max RF, P-value and the reference. The normalized value is formatted to four decimal places before export; it is not a full-precision ratio.
Figures and text artifactsThe two result figures, input fingerprint, selected methods, completed-run status and summary. Native label crowding and clipped sidebar references remain documented visual limits.
Separate manifest and original treesArtifact names and byte hashes establish the saved bundle’s identity. Keep both original trees separately: these summary tables do not export complete split sets or a full copy of each input tree.

The raw integer distance and denominator remain available in the saved normalized table, so preserve them when interpreting rounded values. A displayed four-decimal match alone is weaker than agreement of those exact fields. File integrity is separate from scientific validity.

Worked examples describe particular inputs, methods and recorded results. Check current availability and input requirements before planning an evaluation.

← Back to the learning center