Phenotype match

[StephanHolgerD]

Hi,
is there an option to see the semsimian similarity in the output?

Somehow its hard to set the config to a reasonable value if there is no reference for good phenotype matches and rather bad matches.

How did you end up with the default value 14 ?

best,
Stephan

[MattWellie]

Hi @StephanHolgerD,

Settling on 14.0 required a lot of experimentation locally using participant phenotypes and diagnostic variants. We iterated on the process and thresholds until we obtained a automated performance level that matched manual specialist analysis. It would be hard to reverse engineer this decision and explain the basis for that number, other than to say the tool achieved really strong results using that threshold value, based on our test data.

I've had a low priority issue to improve this part of the process for a while (and a now terribly outdated PR). Semsimian requires a ton of reference data (the largest file in the whole pipeline) and runs relatively slowly. HPO3 is faster, and installs with default reference data. I've struggled to reconcile the results produced by both tools, even when they claim to use the same underlying comparison model, which has made it hard to move on from Semsimian.

Surfacing the similarity score in the results (JSON) shouldn't be too difficult, explaining the value of a given numerical result (and possibly including the exact matching term/terms which gave that result) will be a bit more work. I'll raise a proper ticket internally to see how easy/possible it is to store some additional data about the match, and to see if that can be explained/presented in a way which makes sense in the report. If the Semsimian result can't be meaningfully explained, it might be more reason to move on from the tool and on to something which is less of a black box for HPO prioritisation.

[StephanHolgerD]

Hello, thanks for the response.

Some kind of score for the phenotypic overlap in the json results or html would be great.

At the moment its not clear how strong the evidence is for a specific variant.

best,
Stephan