Convergent phenotyping

[karafecho]

In a recent meeting with Tony Hickey, external SME, he provided feedback on the Translator UI related to what he refers to as "convergent phenotyping", which is essentially the idea of leveraging shared phenotypes to identify treatments for rare disease, or drug repurposing for rare disease based on common phenotypes or common diseases with established treatments.

One example that Tony provided is that many rare diseases are associated with seizures. What are the commonalities? Are there treatments/therapies that might target those commonalities?

One of the Translator MVP1 results that Tony seemed to like was for an inferred response that included 26 paths linking a drug to a rare disease by way of one or more phenotypes. He suggested that it would be helpful for users to see a cross-section of genes among those paths, i.e., grouping the paths by intermediate genes.

Other specific (somewhat redundant) suggestions included the following, some of which have been suggested by other SMEs:

• Suggest an explorer function to allow users to map out from node(s), e.g., phenotypes, seek nearby nodes and connections; support user exploration of subgraph spaces
• Show me other paths that might connect gene A to disease B
• What drugs may treat phenotypes of a rare disease? Or, given a set of phenotypes for a given rare disease or common phenotypes among multiple rare diseases, are there commonalities that may lead to the identification of a drug for repurposing?
• Show other indications for a given drug, group results by indication
• Allow users to explore the graph nearby the actual answer subgraph or path

Note that the answer coalescer / graph enrichment feature that ARAGORN developed may address at least some of Tony's comments and desired features.

[gprice1129]

@karafecho this definitely requires some discussion and breakdown.

[mbrush]

I had a conversation with @cmungall at the Relay that may be pertinent to this issue, about incorporating results from the Monarch phenotypic similarity tool into Translator.

Monarch has a tool that can quantify the similarity between two entities (e.g. two different Diseases, a Gene and a Disease) based on the similarity of their phenotype profiles (the set of all phenotypes associated with them, in KBs such as HPOA). See the slide here for an example that shows a phenotypic comparison between two rare disease patients with related phenotypic profiles.

In theory, we could use this Monarch tool to pre-compute phenotype similarity between any two diseases, or disease-gene pairs - and define a score cutoff above which we create an 'is phenotypically similar to' edge between them. We could then stand up a KP to serve these edges for use by ARAs performing reasoning/prediction.

This could provide valuable input for algorithms looking for drug repurposing opportunities (if Drug X treats Disease Y, it may also treat phenotypically similar Disease Z). Longer term, we could think about how to more dynamically incorporate these types of tools from Monarch into Translator (e.g. a workflow to allow dynamic query of a user input set of phenotypes to find phenotypically-similar diseases, and look at known treatments for those diseases).

[Mathelab]

Quick note also that Gwenlyn also has a hashing algorithm that can compute similarity between any entity types. Not sure how to tag her here but this is very relevant as well.

[karafecho]

@gglusman : See above.

[karafecho]

@gprice1129 : Apologies for packing this ticket with multiple issues. Sometimes that's the easiest way to make sure I've captured and documented issues that are generated during a SME engagement session, but I recognize the downstream impacts. I guess I would say that the main issue here is the one that is receiving the most attention, namely, the general concept of "convergent phenotyping" and leveraging shared phenotypes to identify drug treatments for rare disease. The other issues, e.g., explorer function, are largely covered by other tickets, I think.

Hope this helps ...

[gprice1129]

@karafecho I think its perfectly reasonable to pack one ticket with several issues, but at some point a breakdown of that ticket needs to happen to spawn actionable tickets. When that happens we can close this one.

[sierra-moxon]

TAQA task: break down this issue.

[sstemann]

Other specific (somewhat redundant) suggestions included the following, some of which have been suggested by other SMEs:

1 Suggest an explorer function to allow users to map out from node(s), e.g., phenotypes, seek nearby nodes and connections; support user exploration of subgraph spaces 2 Show me other paths that might connect gene A to disease B 3 What drugs may treat phenotypes of a rare disease? Or, given a set of phenotypes for a given rare disease or common phenotypes among multiple rare diseases, are there commonalities that may lead to the identification of a drug for repurposing? 4 Show other indications for a given drug, group results by indication 5 Allow users to explore the graph nearby the actual answer subgraph or path

Note that the answer coalescer / graph enrichment feature that ARAGORN developed may address at least some of Tony's comments and desired features.

I numbered these and believe

1 - next phase 2 - seems like pathfinder would be a good first step 3 - tangential to MCQ 4 - indications still seem difficult. O&O is tackling similar functionality with ATC Codes 5 - next phase