concept · intermediate · partly written
How a phenotype ontology turns features into a ranked diagnosis, and the evidence
Phenotype similarity plus variant scoring ranks candidate diagnoses; the measured yields come from solved cases, and no outcome study was found.
Also called: Phenomizer, Exomiser, phenotype-driven variant prioritisation
The four steps
A clinician writes findings as HPO terms. A similarity score weights rare shared terms more than common ones. A ranker such as Phenomizer applies it to the query and each disease. In genome work a phenotype score is combined with variant rarity and predicted damage (Exomiser) and a person reviews the ranked list. 12
Reported, not yet replicated
Information content more
Philip Resnik defined, for word taxonomies in 1995, the information content of a concept as the negative log of how often it or its descendants occur, and the similarity of two concepts as the information content of the most informative concept above both. His abstract reports a correlation of 0.79 with human similarity judgments, against 0.66 for edge counting. Sharing a rare ancestor counts for more than sharing a common one. 3
Reported, not yet replicated
Phenomizer and Exomiser more
The 2009 Phenomizer paper adapted such metrics to compare a set of HPO terms with each disease and added p-values; its abstract reports that it beat term matching that ignores term relations, most for noisy queries, with no accuracy figure in the abstract. The Exomiser ancestor PHIVE compared mouse-model phenotypes (28,176 models) using OWLTools. 24
Reported, not yet replicated
Results on simulated and solved cases more
On 100,000 simulated exomes built by adding one of 28,516 known disease mutations to 1000 Genomes exomes, the correct gene ranked first in 83% (recessive) and 66% (dominant), against 77% and 28% for variant-only ranking. On 134 solved retinal exomes with default settings, the diagnosed variant ranked first in 74% with HPO terms and 3% without; the top five fell from 94% to 27%. On 4,877 solved 100,000 Genomes cases Exomiser ranked the diagnosis first in 82.6% and in the top 10 in 93.6%. 451
Reported, not yet replicated
Yield in practice more
In the 100,000 Genomes pilot, 25% of probands were diagnosed (2,183 families), 35% for likely-monogenic disorders against 11% for likely-complex; this is whole-pipeline yield, not the ranker's share. In a prospective 40-patient study, 11 were diagnosed (28%) at mean rank 2.4, one centre, abstract only. 67
Reported, not yet replicated
Where it breaks more
Benchmarks are simulated from the same knowledge the tool uses, or are cases already solved, so high recall shows the machinery works, not that it finds new answers. About 5% of known diagnoses were not detected at all in the 100,000 Genomes benchmark (1% low-quality variants, 2% high allele frequency, 3% incomplete penetrance), and 75% of probands stay undiagnosed after state-of-the-art analysis. The 74% rests on one fixed term set per diagnosis in one disease area. 15
Reported, not yet replicated
A disputed benchmark more
An independent benchmark with default settings put Exomiser top-ranked in 15% of 305 cases from one cohort and 26% of 209 in-house cases, against 38 to 58% for other tools. The developers replied that with recommended settings the figure on 161 solved singleton cases rose from 34% to 72%. The reply is by the tool's developers; a later benchmark paper by developers says the independent group's shared data helped resolve the discrepancy. This atlas found no independent re-run of the 72% on public data. 8910
Contested
Language models against the ranker more
On 5,213 solved case reports, the best of seven language models ranked the correct diagnosis first in 23.6% of cases, against 35.5% for Exomiser using phenotype only. Only the abstract was read. 11
Reported, not yet replicated
What was searched and not found more
No randomised or prospective study comparing phenotype-driven ranking with no ranking in new patients was found, and no study in which an ontology-based diagnosis changed treatment, survival or cost. No head-to-head comparison of Phenomizer and LIRICAL was found. Vendor claims of higher yield were left out as unsourced. 61
Reported, not yet replicated
Plain answer
Why it matters
It is the strongest clinical case for ontologies in this atlas, and it shows the limits of the evidence: solved-case rank, not patient outcome. 1
Our inference
Real numbers
Top-1 on 134 retinal exomes, with and without HPO terms
74% with HPO terms and 3% without, default settings. 5
Reported, not yet replicated
Overall diagnostic yield, 100,000 Genomes pilot
25% of probands diagnosed. 6
Reported, not yet replicated
Still open here
- Does a representation-only comparison exist for diagnosis ranking (HPO terms against free text with the same ranker)?
- Has the 72% figure been reproduced by an independent group on public data?
Not yet written: Equation.
Linked from
Sources
- Jacobsen JOB, Kelly C, Cipriani V, Research Consortium GE, et al.. Phenotype-driven approaches to enhance variant prioritization and diagnosis of rare disease. Human mutation (2022). paper; opened 2026-11-18 all records citing this
- Köhler S, Schulz MH, Krawitz P, Bauer S, et al.. Clinical diagnostics in human genetics with semantic similarity searches in ontologies. American journal of human genetics (2009). paper; opened 2026-11-18 all records citing this
- Philip Resnik. Using Information Content to Evaluate Semantic Similarity in a Taxonomy. arXiv (1995). preprint, not peer reviewed; opened 2026-11-18 all records citing this
- Robinson PN, Köhler S, Oellrich A, Sanger Mouse Genetics Project, et al.. Improved exome prioritization of disease genes through cross-species phenotype comparison. Genome research (2014). paper; opened 2026-11-18 all records citing this
- Cipriani V, Pontikos N, Arno G, et al.. An Improved Phenotype-Driven Tool for Rare Mendelian Variant Prioritization: Benchmarking Exomiser on Real Patient Whole-Exome Data. Genes (Basel) (2020). paper; opened 2026-11-17 all records citing this
- 100,000 Genomes Project Pilot Investigators, Smedley D, Smith KR, Martin A, et al.. 100,000 Genomes Pilot on Rare-Disease Diagnosis in Health Care - Preliminary Report. The New England journal of medicine (2021). paper; opened 2026-11-18 all records citing this
- Zemojtel T, Köhler S, Mackenroth L, Jäger M, et al.. Effective diagnosis of genetic disease by computational phenotype analysis of the disease-associated genome. Science translational medicine (2014). paper; opened 2026-11-18 all records citing this
- Yuan X, Wang J, Dai B, Sun Y, et al.. Evaluation of phenotype-driven gene prioritization methods for Mendelian diseases. Briefings in bioinformatics (2022). paper; opened 2026-11-18 all records citing this
- Jacobsen JOB, Kelly C, Cipriani V, Robinson PN, et al.. Evaluation of phenotype-driven gene prioritization methods for Mendelian diseases. Briefings in bioinformatics (2022). paper; opened 2026-11-18 all records citing this
- Bridges Y, Souza V, Cortes KG, Haendel M, et al.. Towards a standard benchmark for phenotype-driven variant and gene prioritisation algorithms: PhEval - Phenotypic inference Evaluation framework. BMC bioinformatics (2025). paper; opened 2026-11-18 all records citing this
- Reese JT, Chimirri L, Bridges Y, Danis D, et al.. Systematic benchmarking demonstrates large language models have not reached the diagnostic accuracy of traditional rare-disease decision support tools. European journal of human genetics : EJHG (2026). paper; opened 2026-11-18 all records citing this
Where it sits on the map
History of this record (6)
This record has been changed 6 times, on the dates below. The note kept against each change is written for the people who maintain the atlas, in their own shorthand, so it is not shown here; it is in the record itself.
- Status
- partly written
- Last updated
- 2026-11-19
- Record
- content/concepts/phenotype-ontology-diagnosis-ranking.json