Helena

Scientific briefing

Phenotype matching in genomic variant interpretation

How HPO and ontology-aware semantic similarity connect clinical findings with gene-associated phenotypes and create a reviewable order of genomic evidence.

Helena Bioinformatics  |  Reviewed 8 August 2026  |  22 minute video

22:42 / Scientific briefing with optional English captions

Read the edited transcript

Clinical context

Sequencing finds variants. Phenotype gives them patient-specific meaning.

A genome can contain millions of observations. Technical filtering narrows the list, but it does not answer which candidates fit the person being evaluated. Phenotype matching compares a reviewed patient profile with phenotype knowledge associated with candidate genes.

The comparison works through HPO concepts and their relationships. It can preserve useful similarity when clinical records use synonyms or different levels of specificity. The score helps order review. It does not determine causality or diagnosis.

Phenotype as data

HPO separates clinical meaning from wording

A clinical note is written for people. Computation needs a stable representation of the finding. The Human Phenotype Ontology assigns identifiers to phenotypic abnormalities and places them in a directed hierarchy. A specific finding can therefore remain related to broader findings even when two sources use different labels or levels of detail.

This structure does not repair a poor clinical profile. The selected terms still need to be accurate, current, sufficiently specific and reviewed by a qualified user. Age, disease stage, negated observations and missing findings can change the signal available to the comparison.

Semantic comparison

Profiles are compared through the ontology, not through word overlap

01

Encode the findings

Represent reviewed observations with stable HPO identifiers and retain their case context.

02

Compare profiles

Use ontology relationships and concept specificity to compare the patient profile with gene-associated phenotypes.

03

Organise review

Read phenotype relevance beside classification, inheritance, frequency and molecular consequence.

Folklore workflow

The phenotype signal stays attached to the evidence path

Folklore stores a reviewed set of HPO terms with the analysis session. Candidate genes carry phenotype profiles derived from curated knowledge. The system compares distinct profiles and presents gene-level relevance beside the variants in that gene. Molecular evidence remains variant-specific.

Gene summaries can arrive before every variant detail is opened. The reviewer can then inspect the classifications, inheritance, frequencies, consequences, sources and term relationships behind a candidate. This summary-first interaction reduces repeated work without turning the score into an unexplained conclusion.

A match score is evidence. It is not a diagnosis.

Phenotype matching changes the order of attention. It cannot prove causality, repair an incomplete phenotype profile or override contradictory molecular and inheritance evidence. A qualified clinical geneticist interprets the complete record.

Interpretation boundary

Where phenotype matching can fail

  • Incomplete profile. A young patient may not yet express the full phenotype, and an omitted finding can lower a relevant candidate.
  • Annotation gaps. Gene and phenotype knowledge changes. The configured reference release may not contain a genuine relationship.
  • Broad findings. Common ontology concepts carry less discriminatory information than specific observations.
  • Phenocopies and conflicting evidence. Similar presentations can arise through different mechanisms. Inheritance and molecular evidence can contradict an attractive semantic match.

Primary scientific sources

  1. 1. Gargano MA et al. The Human Phenotype Ontology in 2024. Nucleic Acids Research. 2024;52:D1333 to D1346. doi:10.1093/nar/gkad1005
  2. 2. Köhler S et al. Clinical diagnostics in human genetics with semantic similarity searches in ontologies. American Journal of Human Genetics. 2009;85:457 to 464. doi:10.1016/j.ajhg.2009.09.003
  3. 3. Jacobsen JOB et al. Phenotype-driven approaches to enhance variant prioritization and diagnosis of rare disease. Human Mutation. 2022;43:1071 to 1081. doi:10.1002/humu.24380
  4. 4. Human Phenotype Ontology. Official resource

Accessible text version

Edited video transcript

Pause markers, repeated spoken phrases and verbal corrections have been removed for reading. The scientific meaning and sequence follow all 23 slides. The video includes optional English captions that can be turned on or off in the player.

  1. 01

    Phenotype matching

    Welcome to our presentation on phenotype matching. We will look at how clinical findings are encoded with HPO, how ontology-aware semantic similarity compares patient and gene profiles, and how Folklore places phenotype relevance beside variant and case evidence. Sequencing can produce a large set of technically credible variants. The phenotype supplies patient-specific context for ordering their review. The result is an ordered set of genes and variants for qualified review. A phenotype match changes the order of attention. It does not establish causality or supply a diagnosis. The clinical geneticist remains responsible for interpreting the complete evidence record.

  2. 02

    The interpretation bottleneck

    Sequencing is very good at producing observations. A whole-genome workflow may begin with millions of variant records and still leave the central clinical question unanswered: which observations could plausibly explain this patient? Technical filtering reduces the search space, but it does not supply patient-specific meaning. Phenotype matching asks whether the clinical pattern observed in the patient resembles the phenotypic pattern associated with a candidate gene or disease mechanism. The result is a more useful order of attention. Candidates that fit the case can move toward the top of the review queue, while weakly aligned candidates remain visible.

  3. 03

    Phenotype as data

    A phenotype is an observable characteristic: a seizure type, a pattern of growth, a structural anomaly, a laboratory abnormality or another clinically meaningful finding. A phrase in a note is not automatically computable. The Human Phenotype Ontology provides a persistent identifier, synonyms and a position in a hierarchy. That structure separates meaning from wording. A clinician may use a familiar phrase, a publication may use another expression and a database may store an HPO identifier. Once those expressions resolve to the same concept, they can participate in a reproducible comparison. Moving from narrative observation to a reviewed, coded finding is the first technical boundary of phenotype matching.

  4. 04

    The Human Phenotype Ontology

    The Human Phenotype Ontology represents phenotypic abnormalities as a directed hierarchy. A specific term such as generalized tonic-clonic seizure sits below broader concepts such as seizure, nervous system abnormality and clinical abnormality. Clinical records are rarely written at identical levels of specificity. The ontology makes relationships between broad and specific findings explicit. It also supports synonyms and structured disease annotations, allowing software to reason over clinical meaning rather than literal words. Folklore uses this shared vocabulary as the semantic layer between patient findings and gene-associated phenotype profiles. The platform applies it consistently, preserves provenance and keeps the result reviewable.

  5. 05

    Why exact matching fails

    Exact string matching fails for three common reasons. Synonyms may describe the same observation with different wording. One source may record a broad parent concept while another records a specific child term. A patient phenotype can evolve with age, and a gene annotation may reflect only what has been reported so far. A useful system therefore needs the ontology graph, so it can recognise broader, narrower and related findings while preserving the distinctions between them. Semantic methods do not make missing evidence disappear. They make the available evidence less brittle when records contain noise or imprecise terms.

  6. 06

    Semantic similarity

    Semantic similarity compares concepts through their position in the ontology and through how informative those concepts are. Two profiles may share no identical labels and still meet at a meaningful ancestor in the HPO graph. A broad concept such as abnormality of the nervous system is common and contributes limited discrimination. A rare, specific shared concept carries more information. The system evaluates each patient term against the candidate gene profile, finds the most informative counterpart and then considers the profile as a set. The comparison is based on structured meaning, not a count of matching words or a language model judgment that two phrases sound similar.

  7. 07

    Input quality

    The quality of the phenotype profile determines the quality of the signal. Specific terms are valuable when they are supported by observation. False precision is not. Age and disease stage matter because some findings may not have appeared, while others may no longer be prominent. Negated findings can be useful when captured and interpreted carefully. Clinical notes can support term discovery in Folklore, but the selected HPO profile remains a reviewed input. A qualified user confirms what belongs in the case. A useful profile is accurate, current, sufficiently specific and traceable to the case record.

  8. 08

    Folklore workflow

    The patient clinical profile is encoded as reviewed HPO terms. Candidate genes carry phenotype annotations derived from curated reference sources. The system compares the patient profile with each distinct gene profile using the HPO graph. Phenotype relevance is then read alongside variant classification, consequence, frequency, inheritance and other case-specific evidence. The output is organised into reviewable priorities and delivered with the underlying evidence accessible. The phenotype signal remains attached to the case. The user can move from a ranked gene summary to variants, classifications, sources and individual term relationships that support the ranking.

  9. 09

    Patient profile

    The workflow begins with observed features. A clinical narrative may contain symptoms, onset, severity and temporal context. The application helps turn relevant positive findings into stable HPO identifiers, and the reviewed selection is stored with the analysis session. The original clinical language provides context, the identifiers provide a computable representation and the session record preserves which terms were used. Careful curation may revisit terms when new clinical information appears. The profile is the current authorised representation of the case. Treating it as data with provenance makes later comparisons reproducible and helps the reviewer understand a candidate score.

  10. 10

    Candidate profile

    Genes are associated with phenotype profiles assembled from curated disease and phenotype knowledge. Variants in the same gene can share a biological phenotype context even though their molecular consequences differ. Folklore canonicalises the gene and HPO profile and avoids recomputing the same semantic comparison for every variant in the gene. The gene-level result can be attached to relevant variants while variant-specific evidence remains distinct. This reuse does not imply that every variant in a phenotype-aligned gene is causal. Molecular evidence still requires review.

  11. 11

    Set-to-set comparison

    The comparison is set-to-set. For every patient term, the system searches the candidate gene profile for its most informative semantic counterpart. Those term-level relationships are combined into a normalised profile-level signal. A single attractive match does not automatically dominate an otherwise poor profile. Broad terms contribute less discrimination than specific terms. Repeated or equivalent gene profiles are deduplicated before computation. The result can be traced to individual term relationships. The platform can present both the summary score and the clinical concepts that created it.

  12. 12

    Whole-genome scale

    Whole-genome analysis makes computational efficiency part of clinical usability. Folklore first canonicalises term sets so equivalent profiles have a stable representation. It deduplicates repeated gene profiles, uses cached ontology relationships and reuses completed comparisons. Compact gene summaries are prepared first and variant detail is loaded when requested. This avoids making a geneticist wait for every row of every variant before useful review can begin. It also preserves separation between durable clinical records, reference knowledge and high-volume session analytics.

  13. 13

    Interpretation boundary

    A phenotype match score is evidence, not a diagnosis. It can focus attention, reveal that a candidate gene resembles the patient clinical profile and help organise a large result set. It cannot prove that a variant is causal. It cannot compensate for a missing phenotype, an incorrect annotation or a gene to disease relationship absent from current reference knowledge. It cannot replace assessment of zygosity, inheritance, penetrance, segregation or molecular mechanism. Folklore keeps phenotype relevance as one visible component of the record. A high semantic score must not be mistaken for a final interpretation.

  14. 14

    Integrated priority

    Folklore keeps several evidence layers distinct. Phenotype fit asks whether the gene profile resembles the case. Variant class reflects the ACMG and AMP evidence state and resulting classification. Variant context includes consequence, impact, population frequency and transcript information. Case context includes inheritance, genotype, family structure and other available evidence. These layers answer different questions. A strong phenotype fit makes a candidate interesting, but a common benign variant does not become pathogenic because its gene matches the phenotype. A pathogenic finding outside the presenting phenotype may remain important in a different review path.

  15. 15

    Clinical organisation

    Folklore groups results into clinical priority tiers to make a large analysis reviewable. Production thresholds and rules remain controlled. Tier 1 contains high-priority phenotype-aligned findings. Tier 2 contains strong candidates requiring focused review. An incidental path preserves relevant classifications outside the presenting phenotype. Lower tiers retain weakly aligned or lower-priority findings. The tiers organise review. They are not diagnostic categories. Full evidence remains available below the tier label, and changes in phenotype, inheritance or classification can alter priority.

  16. 16

    Case context

    Expected inheritance can strengthen or weaken a candidate depending on genotype and family information. Strong benign evidence prevents an attractive phenotype match from dominating the record. Incidental findings require a distinct path because clinical relevance is not identical to similarity with the presenting phenotype. No single signal receives unlimited authority. Phenotype relevance, classification and inheritance interact, but each remains inspectable. Folklore applies controlled rules to organise those signals. Qualified review remains necessary because penetrance, variable expressivity, mosaicism, phenocopies and incomplete family data can change the meaning of a result.

  17. 17

    Summary-first experience

    Gene summaries arrive first, ordered by review priority and phenotype relevance. A reviewer can see the gene, tier, number of variants and relationship to the patient profile without loading every detailed record. Opening a gene requests its variants and reveals the molecular evidence. Individual phenotype matches can be inspected when needed. This interaction presents a reviewable shortlist before variant detail while keeping the underlying variants available. The scientific hierarchy runs from patient profile to gene relevance and then to variant evidence.

  18. 18

    Fictional case walkthrough

    A fictional patient has progressive spasticity, delayed motor development and gait disturbance. No patient data are used. After the reviewed HPO profile is compared with gene annotations, candidates associated with a coherent spasticity and motor-development phenotype rise in the order of review. The geneticist inspects variants, inheritance, classification, frequency, transcript consequence and sources. A candidate with excellent phenotype fit but incompatible inheritance may fall away. Another with a weaker semantic match but stronger molecular and family evidence may become more important. Phenotype matching changes the order and efficiency of review while leaving the conclusion to the complete record.

  19. 19

    Failure modes

    An incomplete profile can under-rank the correct gene, especially in a young patient whose full presentation has not emerged. Annotation gaps can hide genuine relationships because gene and phenotype knowledge changes. Broad findings carry limited discriminatory value. Phenocopies can produce similar patterns through different mechanisms. Incorrectly selected HPO terms can mislead the comparison. A reviewable workflow supports revision, shows individual matches and records reference versions. Absence of a match is not proof against causality. The result describes the current relationship between the patient data available today and the curated knowledge in the configured reference release.

  20. 20

    Human review

    Folklore can organise evidence, perform repeatable comparisons, rank attention and make the path from phenotype to candidate visible. The geneticist interprets the patient, evaluates limitations, reconciles conflicting evidence and authorises conclusions. Computation contributes scale, consistency and traceability. Human expertise addresses context, uncertainty and clinical consequence. The aim is to give the reviewer a clearer and more reproducible evidence path through a complex genomic analysis.

  21. 21

    Helena Bioinformatics

    Helena Bioinformatics brings research, software engineering and clinical genomics into one system. Folklore is the evidence-centred workspace where those capabilities meet laboratory review. The platform connects versioned reference data, controlled processing and phenotype-driven prioritisation with variant evidence. Phenotype matching sits beside annotation, classification, population data, inheritance and source provenance. It is presented as part of a controlled workflow in which the question, evidence and reviewer remain linked.

  22. 22

    What to remember

    Phenotype matching makes genomic evidence case-specific. Clinical findings are encoded with stable HPO concepts because free text alone is not enough for reproducible comparison. Meaning is compared through the ontology rather than identical wording. Phenotype relevance is interpreted beside variant classification, molecular consequence, inheritance and the rest of the case evidence. Ranking focuses review. It does not replace review. In Folklore, the path remains visible from the reviewed phenotype profile through gene-level relevance to the variants and sources the geneticist assesses.

  23. 23

    Selected references

    The scientific framework is grounded in the Human Phenotype Ontology, clinical semantic similarity research and phenotype-driven approaches to variant prioritisation. The references below provide the primary scientific sources and official documentation used for this briefing. More information about Helena Bioinformatics and Folklore is available on the official websites.

Continue with Folklore