Delineate types of clinically significant variant-condition annotations

[mbrush]

Here we consider the types of higher-order clinical significance annotations that link 'variants' to 'conditions' to which they might contribute or impact clinical management of, and which we might want to define/distinguish in our model. Below is the current proposal and several considerations for settling on a final classification.

Note that the notion of a 'condition' here subsumes 'diseases' (e.g. 'diabetes mellitus type II') and more fundamental 'phenotypic features' (e.g. 'elevate blood glucose', 'obesity'). Although there has been discussion of generalizing even further to 'trait').

Note also that for the purposes of classifying VA types below, we use 'variant' to cover all forms/levels of genetic variation, including simple alleles, CNVs, structural variants, halpotypes, and genotypes.

[mbrush]

Proposal to define an set of 'higher-order' VA categories linking variants to conditions:

Clinical Variant-Condition Annotation
    Causal Variant-Condition Annotation
        Genotype-Phenotypic Feature Annotation
        Variant Pathogenicity Interpretation
        Variant Tumorigenesis Annotation
        Variant-Side Effect Annotation
    Diagnostic Variant-Condition Annotation
    Prognostic Variant-Condition Annotation
    Condition Risk Variant Annotation
        Polygenic Risk Variant-Condition Annotation
Variant-Treatment Response Annotation (aka 'Predictive', 'Therapeutic')

Definitions:

Clinical Variant-Condition Annotation: makes a clinically significant statement about an association between a variant and a condition.

• Causal Variant-Condition Annotation: makes a statement about a causal association between a variant and a condition, wherein the variant may contribute to or independently cause the condition to develop.
  • Variant Pathogenicity Interpretation: makes a statement about a causal association (or lack thereof) between a variant and a Mendelian condition, wherein the variant is described along a spectrum from benign to pathogenic (e.g. via ACMG 2015 guidelines).
  • Variant Tumorigenesis Annotation: makes a statement about the contribution a variant makes to the pathogenesis of a particular type of tumor (e.g. if it is a driver, modifier, or passenger variant).
• Diagnostic Variant-Condition Annotation: makes a statement about a pathognomonic association between a variant and a particular type of condition (i.e. whether the variant is specifically characteristic or indicative of the condition such that it allows for a clear diagnosis).
• Prognostic Variant-Condition Annotation: makes a statement about the correlation between a variant and the progression, severity, or outcome of a particular type of condition (following its onset).
• Condition Risk Variant Annotation (formerly 'Predisposition'): makes a statement about the likelihood of developing a particular disease prior to its onset, given the presence of a particular genetic variant (e.g. high, medium, low, or no risk).
  • Polygenic Risk Variant-Condition Annotation: makes a statement that predicts susceptibility to a polygenic disease based on aggregating risk effects of multiple variants into a single genetic risk score.

Variant-Treatment Response Annotation: an annotation about a variant as a predictor of a particular response to a treatment in the context of a particular condition, including therapeutic effect of the treatment (sensitivity vs resistance), pharmacokinetics and toxicity of drug processing, or specific side effects (e.g. leukopenia).

• Variant-Side Effect Annotation: makes a statement about a variant as a predictor of a particular side-effect in response to a treatment in the context of a particular condition.

[mbrush]

*Note that what we define as a 'Predisposition' annotation may be different what CIViC means by this term, where the outcomes of their 'Predisposing' interpretations seem equivalent to Variant Pathogenicity Interpretations of germline variants - asserting whether a variant is pathogenic or not for a particular condition. For example see the CIViC 'Predisposing' assertion record here, which clearly describes pathogenicity as defined using the ACMG guidelines.

By contrast, what we aim to describe in our 'Predisposition' VA type are statements about the probability or level of risk for developing some condition associated with a particular variant - which are typically framed as qualitative categories (high, medium, low), or quantitative genetic risk scores. Such assertions are based primarily on statistical evidence such as odds ratio or relative risk calculations. This is different than Variant Pathogenicity Interpretations or CIViC Predisposing assertions, which are based on more complex and diverse types of evidence (e.g. as defined int he ACMG Guidelines), and make a different type of claim about whether the variant is causal for a condition or not. Obviously these are closely related, but from the perspective of the types of annotations/statements we have observed in the literature and databases, it seems worth distinguishing these initially.

Given the inconsistent use of 'Predisposition' in different communities/systems, consider if we might rename this VA type to something like “Variant-Condition Risk Annotation”, or “Quantitative Risk Annotation”. This would keep the same basic definition (an annotation about the likelihood of developing a particular condition prior to its onset, given the presence of a particular variant), but provide a label that is more explicit about what is stated in this VA type.

---

The Tumorigenicity VA type is another one that I am not very familiar with, and it is unclear if and how this should be distinguished from its proposed Pathogenicity parent type. It has been distinguished initially because there seem to be different classifications of variants in cancer that describe if/how they contribute to a tumor (e.g. driver, modifier, passenger), and are presumably cancer-specific guidelines for classifying variants in this way. Hoping whoever takes a deep dive into this VA type can provide further insight, and weigh in on if/how this shoudl be distinguished from other VA types.

[mbrush]

One area to be sure and sort out any confusion and provide clearly discriminating definitions is between the Pathogenicity vs Diagnostic vs Prognostic VA types. In my mind the definitions above make clear distinctions. 'Pathogenicity' assertions are more mechanistic in nature - stating a variant is causal. Diagnostic and Prognostic assertions are focused on a variant's impact for clinical care (if it can lead to a definitive diagnosis, or suggests a particular outcome). But I'm not sure these align with community thinking/use of these terms, or if being diagnostic or prognostic necessarily requires the variant to be causal/pathogenic or not.

Figure 2 in the ClinGen Somatic WG MVLD paper suggests that the somatic Diagnostic, Prognostic, and Predictive categories may all be subtypes of Pathogenicity interpretations.

This would suggest that asserting a variant has diagnostic, prognostic, or predictive significance for a particular condition assumes that the variant must be pathogenic for the condition as well. Curious if others would agree with this, and the idea of making these somatic interpretation categories subtypes of the Pathogenicity VA type?

[mbrush]

Where the proposed distinction between causal (Variant Pathogenicity) and predisposing (Variant Predisposition) VA types gets blurred is in ClinVar - where variants are asserted as causal for cancer predisposition syndromes. e.g. this RCV that references 'Hereditary breast and ovarian cancer syndrome' that is defined by an "increased risk for female and male breast cancer, ovarian cancer".

Something to keep in mind as we defined and model these VA types, and consider if/how to distinguish them.