[Ontology-editors] Abductive reasoning

[Mike Bada]

I agree with Chris that most of the abductively inferred nonalignments are
probably wrong and that the subject and object terms should be aligned.
(That is, both sides should be subsumptively linked or both should not be
subsumptively linked.) There is a third option, though, and that's to use
a disjunction as the object of the more general assertion. For example:

regulation of transport   regulates   transport
regulation of transporter activity   regulates   transporter activity
regulation of transporter activity   is_a   regulation of transport

This is a nonalignment in that "regulation of transporter activity" is
subsumed by "regulation of transport", but "transporter activity" is not
subsumed by "transport". To align them, you could either remove the is_a
link between "regulation of transporter activity" and "regulation of
transport", or you could add an is_a link from "transporter activity" to
"transport". The third option, which won't align the two assertions but
will make them mutually consistent, is to make the object of the more
general assertion a disjunction:

regulation of transport   regulates   (transport OR transporter activity)

Now, a regulation-of-transport process regulates either a transport process
or a transporter activity, and a regulation-of-transporter-activity process
is further constrained to regulate a transporter activity.

Cheers,
Mike



Chris Mungall wrote:
> This came up on one of the trackers. Most of the reasoning we do (for 
> checking parentage etc) is deductive. However, abductive reasoning has 
> proven useful too. Mike has a paper on this.
>
> The difference between deductive and abductive reasoning is explained 
> nicely here:
>
> http://en.wikipedia.org/wiki/Abductive_reasoning
>
> Deductive reasoning is watertight. The OE reasoner performs deduction. 
> It doesn't do abduction yet. For some of the xps (notably the 
> regulation xps) I perform the additional abductive reasoning step 
> outside oboedit.
>
> Example of deduction. Given 3 assumptions:
>
> [A1] regulation of A = any process that regulates A
> [A2] regulation of B = any process that regulates B
> [A3] A is_a B
>
> The following fact is entailed:
>
> [A4] regulation of A is_a regulation of B
>
> Example of abductive reasoning. The 3 stated facts:
>
> [A1] regulation of A = any process that regulates A
> [A2] regulation of B = any process that regulates B
> [A4] regulation of A is_a regulation of B
>
> then A4 has possible explanation:
>
> [A3] A is_a B
>
> Note that A3 does not *necessarily* hold (to assume so would be a logical
> fallacy as the wikipedia page suggests). However, with GO xps I 
> believe it
> holds enough of the time to make this a useful heuristic (David and Tanya
> can confirm). For example, if the ontology has A1,A2 and A4 but not 
> A3, we can make a pretty good bet that A3 should really be there.
>
> Context:
> https://sourceforge.net/tracker/?func=detail&atid=440764&aid=2087283&group_id=36855 
>
>