Modeling 'Genetic Condition' (as a Domain Entity)

[mbrush]

Creating this ticket to begin collecting requirements and considerations for modeling "Genetic Conditions". This is the name we have given to the concept of "A disease or set of one or more co-occurring phenotypic features, typically controlled by a single gene or locus with a defined inheritance pattern." At the highest level, we should also consider distinctions between concepts such as Disease, Phenotype, Trait - and their relationship to each other and the notion of a Genetic Condition. But here we consider more practical matters.

Specifically, "Genetic Conditions" fill the descriptor slot in many VA types (e.g. Variant Pathogenicity and Variant Oncogenicity Interpretations), and the qualifier slot in others (e.g. Therapeutic Response Interpretations). In these contexts we are concerned with representing "types" of Conditions, as opposed to specific instances that describe Conditions as they manifest in a particular patient.

While we could simply rely on existing disease and phenotype ontologies to provide terms for these descriptors, some driver projects (e.g. ClinGen) have provided compelling use cases for allowing richer and more flexible representations of Genetic Conditions that require defining an object model. The model defined for ClinGen's Variant Interpretation model may serve as a straw man starting point for what this might look like. Requirements from this effort include:

• ability to specify a disease and/or a set of one or more phenotypes that characterize the core condition manifest in a patient.
• ability to capture the inheritance pattern as a key identifying characteristic of the condition
• ability to capture a causal specific gene or genes that characterize the condition.
• ability to define conditions based on temporal progression of phenotypes/manifestations

Again, the goal here is not to model 'instances' of conditions that affect specific patient - so patient-specific features like 'severity' or 'age of onset' may not be relevant (unless they are a distinguishing aspect of a class of disease). The CDPC / Phenopackets work is developing models for describing phenotypes and diseases from this patient-level perspective.

Finally, there are three key subtypes of conditions/diseases that may require specializations to support different types of variant annotations, and serve the needs of different disease research communities:

1. Mendelian Condition: These are caused by single gene, germline mutations. They are relevant for Variant Pathogenicity Interpretations that describe variants that alone are capable of causing disease.
2. Cancer: These are caused by a combination of several somatic mutations, some of which activate oncogenes, others which inactivate tumor-suppressor genes, and likely others that modify/contribute to the progression of the disease in other ways. They are relevant for Variant Oncogenicity Interpretations and other somatic interpretation types.
3. Common Disease: These are multi-genic, with contributions from many variants/genes. They are relevant for Condition Risk and Polygenic Risk annotation types, as these describe variants that may predispose you to getting a disease, but alone may not be 'causal'.

[mbrush]

Some Questions:

• Do we need a proper object model to allow richer descriptions of diseases (as opposed to leveraging simple ontology terms)?
• How to distinguish our 'Condition Type" model from a "Condition Instance" model that would allow description of a particular patients manifestation of disease?
• What requirements do we have for modeling ConditionTypes?
• What other efforts need to model the notion of Condition Type and what are their requirements (in GA4GH and beyond)?

[mbrush]

A straw man proposal based on the SEPIO-ClinGen Variant Pathogenicity Interpretation Model (link) is drawn up in the va-spec modeling spreadsheet here.

[mbrush]

Coordinate with Phenopackets representation of Disease: https://phenopackets-schema.readthedocs.io/en/latest/disease.html . . . but this is a model of a disease instance (i.e. a patient's manifestation of a specific disease), whereas we need to model types of conditions.

[mbaudis]

@mbrush I suggest you directly reference this through its SchemaBlocks representation, and go to Phenopackets if you need upstream changes (which then also would be versioned in {S}[B]).

Nice {S}[B] use case - happy to help!

[mbrush]

We may want to be able to associate a disease or phenotype that characterizes the Condition with its specific inheritance pattern, or associated gene(s). The model as currently specified here would not support this.

An alternative proposal here would - by adding a 'hasComponent' field where a compound Condition could reference more fundamental Conditions that comprise it as nested Condition objects within the larger compound one. Within a given Condition then cardinality on the disease and phenotype fields would become 0..1.

Decision: For v0 go with original simple model, test this, collect feedback/requirements, and evolve to more complex model if needed.