Epistemic formalism (was Re: [Phenoscape] Re: [go] evidence code ontology)

[Larry Hunter]

On Feb 6, 2008, at 3:10 PM, David Hill wrote:
> Until we capture context it is difficult to make a firm conclusion  
> about one reference contradicting another.

David,

While I agree that context is important, I think your question is  
orthogonal to the epistemic relationship question.   My argument is  
three-fold: (1) context can be represented in assertions using the  
ontology itself, not in second-order relationships between hypotheses  
and evidence.  (2) the approach I have been developing represents the  
relationship between evidence and hypothesis, not between different  
evidences; there is no effort to build relationships between  
references. (3) While it may be possible to develop relationships  
among multiple "supports" assertions (or other epistemic relations),  
it is not necessary to do so in order to get the significant benefits  
that having epistemic relations in the RO can offer.

First, context can be represented with existing relations.   You said

> I have thought hard about how as a biologist, I decide when I think  
> experiments hold for all contexts and when they don't. What it comes  
> down to is accumulated knowledge that is based on repeated and  
> various experiments.Contextual information and the variety of  
> contexts influences my opinion as to whether an experiment shows a  
> 'typical' result. Somehow we need to be able to capture this process  
> when we interpret annotations.

First, I would suggest that the issue should not be about  
*interpreting* the annotations, but about *making* them.   
Interpretation of annotation should to the extent possible be  
unambiguous.  If the evidence in an article supports a conclusion only  
in a very narrow context, the resulting annotation should capture  
exactly that.  That's the job of the annotator, to use his/her  
accumulated knowledge to represent the real meaning of the article.   
But the existing ontology and RO (or at least soon to be versions) are  
perfectly capable of doing this without any epistemic elaborations.   
Perhaps the best illustration of this phenomenon was raised by David  
Botstein at the SAB meeting in September: mouse Cytochrome C is  
annotated with both functions ‘electron transport’ and ‘positive  
regulation of apoptosis’ and both locations ‘mitochondrion’ and  
‘cytosol.’ However, these activities and locations have a specific  
interaction: the function is ‘electron transport’ in (and only in) the  
‘mitochondrion,’ and ‘positive regulation of apoptosis’ in (and only  
in) the ‘cytosol.’  In the SAB report, we strongly recommended getting  
cross-products into the annotation pipeline so this sort of thing  
could be handled properly, without having to dramatically expand the  
GO (e.g. to have positive-regulation-of-apoptosis-when-in-the- 
cytoplasm).  Important contextual information should be captured in  
the annotations themselves.

Second, the epistemic relations I have been developing are (only)  
between evidence (a subclass of OBI DENRIE) and a hypothesis (a  
somewhat controversial OBI term, roughly meaning a proposition  
relevant to biology).   I am not proposing to create any RO relations  
between evidences.   There is no logical problem with having instances  
(think annotations) where one supports and the other contradicts some  
particular hypothesis.   You mentioned two possibilities: one kind of  
evidence is better (e.g. more sensitive), or in fact both are true and  
the hypothesis needs refinement (e.g. X is located in Y in cell type  
A; X is not located in Y in cell type B).   The latter can be handled  
as discussed above; the former cannot be represented using the  
existing RO, nor would the epistemic relations I am developing provide  
a method of saying it.

Finally, while we can ultimately come up with relations among  
epistemic relations, we don't need to in order to get significant  
benefit from the basic relations. If we used explicit relations in  
annotations (like "participates in" as I mentioned in my reply to Sue)  
and explicit epistemic relations (instead of evidence codes) to link  
such assertions to evidence, at the very least we can identify the  
different kinds of conflicts, and much more easily provide all the  
evidence relevant to even complex hypotheses using technology like  
HyBrow (hybrow.org).

It might be possible to also develop useful third order relations that  
relate epistemic relations, after we get basic second order ones right.
The Schum paper has a taxonomy (section 4.1.2, p. 29) of evidence  
combination  types, which are broadly grouped into harmonious (or  
mutually supportive) and dissonant.  In Schum's taxonomy, each of  
these is further subdivided into evidence about the same thing (which  
he calls corroborative or contradictory, depending on the direction)  
and evidence about different things that suggest (or don't) the same  
conclusion (which he calls convergent or divergent).  Once we get the  
right basic set (supports, contradicts, is-about, supports-via-<ECO>,  
etc.) then we can look at consonant-with, dissonant-with,  
corroborates, etc.

So, in conclusion, important as context is, I don't think we have to  
address it in order to get the basic epistemic relations into the RO.

Larry