[Phenoscape] Re: [go] evidence code ontology

[Larry Hunter]

On Feb 6, 2008, at 10:23 AM, Peter E. Midford wrote:

> The similarity and homology codes recently added to the ECO on  
> behalf of Phenoscape raise serious issues of contradiction and how  
> to evaluate multiple types of evidence:  evidence from phylogeny  
> (IP) might support or contradict an assertion of homology based on  
> an observable similarity (e.g., IMS).  ...
>
> I didn't raise this issue in the Phenoscape discussion of these  
> terms (directly) because it isn't a reason not to annotate with  
> these codes, but it will affect how we reason with them.

Peter,

You have hit precisely on why I think it is so important that we come  
up with a set of relations to describe the situation.  Even in the  
most basic of assertions regarding biology (e.g. your homology  
example), it is possible that there be conflicting evidence.  It  
should surely be possible for our ontological infrastructure to  
support the formal representation of such situations.

One interesting question, raised indirectly by Matthias Samwald when  
he said:

> http://www.mail-archive.com/public-semweb-lifesci@w3.org/ 
> msg02549.html ->
> scroll down to section "2) What is evidence?", then continue reading  
> the
> replies titled 'Evidence'.

And then in that thread's reference to the work of David Schum (e.g. http://www.evidencescience.org/content/Science.doc) 
  which, if nothing else, clearly demonstrates the complexity of the  
relationship between observations (of various types) and the  
propositions that they are taken to support or contradict.   Although  
Figure 1 (p. 24) is supposed to be a taxonomy of types of evidence, I  
find it informative for thinking about what sorts of relationships  
their might be between observations (of various kinds) and  
hypotheses.   Schum calls it types of "evidence classified on  
inferential grounds."  It is the columns in that figure (the  
"relevance dimension") rather than the rows (the "credibility  
dimension") that is probably more important to the RO (the latter is  
really more about legal evidence than scientific, as all scientific  
evidence is either of the tangible or accepted facts type).  Three  
distinction are made: directly relevant evidence, which can be linked  
to the hypothesis by a "defensible chain of reasoning"; circumstantial  
evidence, which provides "one inferential step... but is always  
inconclusive;" and indirect evidence, which is "meta-evidence" that  
"bears upon the strength or weakness of other evidence."

Do you folks think that it is possible to reliably make a distinction  
between directly relevant evidence and circumstantial evidence based  
on the content scientific publications?  Is it important to do so?

Larry