Statements for assertions repeatedly confirmed by multiple evidence lines?

[fbastian]

Currently, when an assertion is supported by multiple evidence lines, a confidence level can be assigned to the global CI statement, based on the supporting evidence with the highest confidence.

But sometimes, as stated in the CIO paper: "several evidence lines of low quality, produced by methods known to be noisy (e.g., yeast two-hybrid), could yield an assertion of high confidence, because they are repeatedly confirmed (e.g., with different reporter genes)".
There is currently no CI statements allowing to capture this kind of confidence.

Possible solution: create a new CI element of type evidence concordance, like, e.g.: repeatedly congruent.
Should it be subclass of congruent, or directly subclass of concordance of multiple evidence lines?

With this new CI element, we could pre-composed new CI statements, e.g.:
confidence statement from repeatedly congruent evidence lines of same type, confidence high (note how this is not called "best confidence high").
That would be equivalent to:

'confidence statement from multiple evidence lines of same type' 
and ('has evidence concordance' only `repeatedly congruent`) 
and ('has confidence level' only 'high confidence level')

[fbastian]

repeatedly congruent should be assigned based on evidence lines of same experimental or computational method. So, what happens when an assertion is supported by both several evidence lines "repeatedly congruent", and evidence lines of different types?

[marcrr]

In my opinion, this is a call to be made by the resource using CIO, or other CIO users, on what they want to call "high quality" in the end. In this case, you could decide that all "concordance of multiple evidence lines" is high quality, whatever the level of individual evidence lines.

[fbastian]

I partially disagree: we should not complicate the life of annotators, but if you want to say that something is 'high confidence', then you need to use a term with a relation to the appropriate confidence level, not imply it in the dark on your side. Otherwise, there is no standardization possible. And at the very least, it means that the confidences from multiple evidence lines should not have the word "best" in their label.

The question is more: do we want to capture such use cases in a standard way?

Even if we do not provide new pre-composed CI statements, we can provide the ontology mechanisms allowing to be as precise as we want.

[fbastian]

OK, see commit bca02bc

I have modified the labels and definitions of statements from multiple evidence lines associated to a confidence level, so that users are free to choose which strategy they want to apply to define an overall confidence level.

This notably accommodates the "repeatedly congruent" use case, see comments that were added to the ontology. I let the ticket open, so that people can comment about implementing an actual "repeatedly congruent" term.