Clique extracted ontology
Algorithms AIMech:CliXO DRAFT
Aliases: CliXO
An overlapping hierarchical clustering heuristic that constructs a directed acyclic hierarchy from pairwise dissimilarities by selecting informative threshold-graph maximal cliques, with persistence-based noise filtering and optional completion of inferred missing edges.
Identity Notes & Concept Scope
Compared all 442 pinned AIO and 665 native IDs, names, aliases and scoped synonyms for CliXO, its expanded name and publication identifiers; no normalized identity match. Ignored- and hidden-inclusive fixed-string searches of the recovery, older discovery and original Documents checkouts returned 14, 0, 0 lines. All matches were read: only discovery and history notes, with no existing standalone identity. A semantic search including clustering, clique, poset, hierarchy, ontology and community terms covered 41 AIO and 228 native definitions. The 41 AIO and 225 previously read native definitions were unchanged; all three new PoClustering/OHC records were read in full. Eleven complete annotated AIO node/edge/generated contexts and four full native records were assessed. Retain the 2014 informative-clique filtering, persistence and missing-edge-completion heuristic, distinct from exact PoClustering enumeration, greedy edge-cover pruning, OHC density-tolerance expansion and Llama nearest-neighbor unions. Its DAG describes syntactic set hierarchy rather than biological relation meanings. Later code releases remain version-assessment leads.
Claims & Grounded Evidence
7 Claims
Mechanism
#informative-clique-filter
Process dissimilarity thresholds in increasing order and maintain maximal cliques of the resulting graph. The described filter retains a clique when it contains an unexplained pair not covered by previously accepted cliques and a pair unique to that clique among the current clique pool. This differs from retaining every maximal clique or merely pruning an edge-clique cover.
Mechanism
#persistence-noise
A noise parameter delays acceptance until an informative clique persists across a sufficient dissimilarity gap. The paper relates a useful parameter range to separation between true ontology levels and within-term measurement noise; those quantities depend on an unknown target structure, so this is not a universal parameter-selection rule or unconditional recovery guarantee.
Mechanism
#missing-edge-completion
A second parameter controls a minimum normalized vertex-degree condition over unions of overlapping cliques. When the condition is met, the method can lower pairwise dissimilarities to fill inferred missing edges and update the clique pool. The input distance matrix is therefore edited; the parameter is not a learned probability or a guarantee of preserving every original pairwise distance.
Scope
#ontology-structure-scope
The output captures syntactic term sets, their hierarchical relations and assignments of terminal objects to terms, allowing multiple parents and overlapping groups. It does not itself assign biological is-a, part-of or regulates meanings to those relations. Redundant or indistinguishable terms can be invisible in pairwise similarity data, preventing general recovery of an arbitrary original ontology.
Limitation
#printed-specification-boundaries
The printed stack routine leaves next-item lookahead, exhaustion and terminal handling implicit; the strict persistence test also needs a policy for cliques appearing at the last threshold. The imperfect-case height equation uses a different scaling from the perfect-case distance equality; because the imperfect formulation permits enlarged pair distances, this difference alone does not establish a contradiction. Applied simultaneously to a full current pool, the necessary-clique predicate can reject every clique in a complete three-partite graph with two vertices per part. These are specification boundaries, not claims about an uninspected implementation or a general theorem refutation.
Scope
#evaluation-design
The paper separates reconstruction from GO-derived Resnik similarities from experiments using genetic-interaction correlations, gene-expression correlations and YeastNet v3. It acknowledges that YeastNet source weights were trained with GO co-membership. Reported precision and recall use one-to-one ontology alignment with structural constraints and size-binned permutation thresholds; aligned terms are not necessarily identical terms or independently validated biological mechanisms.
Scope
#complexity-and-stability
The stated complexity is output-sensitive and depends on the intended dynamic clique-update procedure; it does not guarantee a small output on all graphs. The stability comparison divides aligned or identical term counts by the smaller ontology size, which can score complete containment despite additional terms in the larger ontology. Parameter-robustness observations are specific to the evaluated networks and settings.
⚠️ Curated Limitations & Negative Results
- This record describes the 2014 Kramer et al. CliXO method. Later software releases and extensions require separate version and mechanism assessment; the author version-0.3 landing page was read only as provenance context.
- The paper supplies a heuristic with noise and missing-edge assumptions, not a certified reconstruction of arbitrary ontologies or a semantic ontology reasoner. Distance editing and invisible terms limit lossless interpretations.
- The static necessary-clique example does not establish how an implementation orders or removes the current pool. Terminal flushing, tie processing and the distance-scaling conventions remain unverified implementation questions.
- The missing-edge example in Section 2.2.2 says k+1, while Section 3.2 correctly gives two k−1 cliques sharing k−2 vertices. Figure 3 legend and prose also differ about the no-completion parameter setting; no disputed numeric result is adopted here.
- All nine primary PDF pages, references, equations, algorithm and figures were read. No author source code, experimental datasets, supplementary material or alternative paper version was audited or executed. Constructed predicate checks are not benchmark replication. This record remains DRAFT.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:HierarchicalClustering
Hierarchical Clustering
|
related_match | CliXO constructs an overlapping hierarchy of object sets. The AIO target denotes the related task, not equivalence to a strict tree or an algorithm/task is-a assertion. |
AIO:UnsupervisedClustering
Unsupervised Clustering
|
related_match | The algorithm groups objects from supplied pairwise relationships. GO-based evaluation and optional domain-informed parameter tuning do not make its generic construction a supervised classifier. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
clixo2014
|
Michael Kramer, Janusz Dutkowski, Michael Yu, Vineet Bafna, Trey Ideker
|
primary paper
full_text
|
ID: doi:10.1093/bioinformatics/btu282
sha256:82c7c90990c1…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-14T00:40:04Z
Updated: 2026-10-14T00:44:17Z
Source File:
data/curated/CliXO.yaml