malachig
commented
6 years ago This all sounds very logical. Some further thoughts.
What do we mean by response? Is that not just a general description of an effect? One that we may wish to describe in more detail? How is an Effect really different from a Side-Effect?
Consider the following contrived examples: A.) Presence of the variant BRAF V600E predicts positive response to dabrafanib in melanoma cancer patients. Where by " positive response" we mean death of tumor cells. For many treatments, we might get more precise: death triggered by apoptosis, by reduced vascularization of the tumor, etc. In non-cancer disease the "response" could be something like reduced risk of heart attack, where again we might get more specific and say reduced blood cholesterol levels, etc.
B.) Presence of the variant DYPD c.1679 T>G predicts adverse response to 5-fluorouracil in colorectal cancer patients. Here by "adverse response" we might mean neutropenia (death of neutrophil cells).
Another way of stating this: Variant X predicts [death of tumor cells] in disease Y when patient gets drug Z. A good thing when treating a cancer patient. Positive response. Variant X predicts [death of neutrophils] in disease Y when patient gets drug Z. A bad thing when treating a cancer patient. Adverse side effect.
Depending on the condition being treated, the adverse side effects could be identical to the positive effects ("response") sought by the use of another drug (or even the same drug) in another context.
In both cases, the disease/condition that I primarily care about is the type of cancer. What changes is whether the treatment effect associated with the variant is a desired "effect" or an undesired "side-effect". This seems a relatively minor (and qualitative) distinction. Hence our decision to model Resistance/Sensitivity and Adverse Responses as different flavors of the same kind of entity in CIViC.
Also, in the proposed Variant Side Effect Annotation, the side effect is the condition. In my example above, neutropenia. Where does colorectal cancer go then?
When an oncologist considers the merits of a particular therapy in patient with a particular kind of cancer, both the positive response and adverse responses may depend on that disease context to some degree. At the very least, they are thinking about them holistically for that disease. A drug they give may have multiple effects some that contribute to positive outcome, others that are unwanted side effects. For both, the starting point is treating a patient with condition X and based on prior experience treating that disease trying to maximize positive effects while minimizing adverse effects.
I think the strategy you lay out above is very reasonable but I'm still wondering about its assumptions. That in Variant Treatment Response the condition is the specific disease (e.g. melanoma) and that thing being predicted is just Response/Sensitivity (not a more specific phenotype). And that in Variant Side Effect the condition is the side effect (e.g. neutropenia, a specific phenotype) and that the disease is a (perhaps less important) context. I think that is all true, but I also think that if we wanted to model Side Effect and Responses in a more unified way, it seems like it would be possible and there would be some benefits to it.
Another thought, researchers working on identifying positive predictors or response are generally quite distinct from those working on predictors or adverse response. If the proposal you laid out is more immediately intuitive to these two communities, that itself has value.
One last thought. In the future one might imagine more widespread adoption of quantitative measures of responses (both for positive and adverse effects). Is that worthy of consideration here?
Anyway, I'm not nearly as versed as you in the ontology, modeling, semantics concepts needed here. I'm satisfied with the current proposal if it helps us move forward. I just wanted to summarize how I was thinking about it when I raised this issue.
mbrush
There has been some debate as to whether Variant-Side-Effect Annotation' fits best under Causal Variant-Condition Annotation, as currently positioned, or under variant-Treatment Response Annotation, (which is about whether a variant is a predictor of response to a type of treatment as used to treat a particular type of condition).
Initially we grouped this under Causal Variant-Condition Annotation, from the perspective that the primary statement it makes is about the contribution of the variant to developing a side-effect (which we consider a condition). From this perspective, the treatment is considered an environmental context that is required for the condition to manifest. In an ACM-based model of this perspective, the descriptor is the condition and the qualifier is the treatment - making the primary statement of the annotation something like "VariantX contributes_to SideEffectY in the context of Treatment Z". i.e. the condition is something that describes a context in which the variant confers a particular response to a drug.
While this view is most intuitive to me, I understand there is history here of wanting to consider Variant-Side Effect annotations with Variant-Treatment Response annotations. And I think this would be possible to do if we extend the predicate set allowable for this annotation type.
Currently we consider roughly the following predicates for Variant-Treatment Response Annotations:
If we extend this lit to include a 'confers side effect in response to' predicate, then we could frame variant-side effect annotations as a subtype of Variant-Treatment Response Annotation with the following schema:
In this case the Condition as a qualifier is refining the meaning of the predicate (from simply stating that the variant "confers side effect in response to" some treatment, to more specifically stating that the variant "confers [qualifier] in response to" the treatment. i.e. the qualifier makes the predicate more specific by indicating the particular side effect that the variant confers in response to the indicated treatment.