Logic and Reasoning Challenges

From phenoscape
Revision as of 18:41, 19 November 2009 by Crk18 (talk | contribs) (Different relations for different purposes)

This page discusses issues to be resolved in the near future. These issues pertain to relation semantics as well as inference procedures.

The problem with absence of features

Descriptions of phenotypes as used in the Phenoscape project (and a plethora of phenomena in the real world) are replete with exceptions, or aberrations from what is considered to be "normal." While canonical ontologies like the FMA and the TAO contain ontological definitions of ideal specimens, observations in the life sciences are full of aberrations to these general rules.

Phenoscape has some typical issues dealing with absence of anatomical features in certain species of Ostariophysian fishes. For example, the basihyal cartilage is found in all species of Ostariophysian fishes, except the Siluriformes. At present, this information is captured in Phenoscape using the combination of the PATO term for "absent in organism" (PATO:0000462), the "inheres_in" relation from the OBO Relations Ontology, the TAO term for "basihyal cartilage" (TAO:0001510), the "exhibits" relation from the PHENOSCAPE ontology, and the TTO term for Siluriformes (TTO:1380). This is shown below.

<javascript> TTO:1380 PHENOSCAPE:exhibits PATO:0000462^OBO_REL:inheres_in(TAO:0001510) </javascript>

In plain English, this translates to "Siluriformes exhibit absence in organism which inheres in basihyal cartilage." The semantics of this sentence are vague to say the least. Going by this methodology, it is impossible to state that basihyal cartilage is absent in Siluriformes without referring to at least one instance of basihyal cartilage. Combining a quality absent with a feature through the inheres_in property is very misleading in itself (ex: absence inheres in cartilage), contorting the intrinsic semantics of the inheres_in relation. These problems have been discussed in Ceusters et al and Hoehndorf et al. Both these publications propose solutions to integrate these aberrant observations with canonical definitions, without causing inconsistencies in reasoning procedures.

Media:PhenotypesInPhenoscape.ppt

Discussion about the Absence of Phenotypes issue

Another issue specific to the Phenoscape project was raised by Paula at the SICB workshop. Given that basihyal cartilage is absent in Siluriformes, basihyal bone should be absent in Siluriformes as well. This is because basihyal bone develops from basihyal cartilage. This may be inferred by adding a new relation chaining rule shown below to the OBD reasoner

Rule:<math>\forall</math>F1, F2, S: absent_in(F1, S) <math>\and</math> develops_from(F2, F1) <math>\Rightarrow</math> absent_in(F2, S)

This relation chain corresponds to the observation GIVEN THAT Basihyal_Cartilage absent_in Siluriformes AND Basihyal_Bone develops_from Basihyal_cartilage, THEN Basihyal_Bone absent_in Siluriformes. This and other similar relation chains (as per identified requirements) are to be implemented for the Phenoscape project in the future. Strategies to deal with absent features in general are also to be implemented in the near future.

Differences between the existing semantics and desired semantics of the exhibits relation need to be resolved to address this issue. Potential strategies to implement the absence of features problem are discussed here.

Inferring in both directions on the taxonomy

It is desired that annotations to higher taxa in the taxonomy be propagated to the lower taxa that are subsumed by the higher taxon; i.e. classical top down inferences. Given that the reasoner already reasons bottom upward, associating phenotype annotations from the lower level taxa to the higher level taxa, adding top-down inferencing may cause widespread inconsistencies in the data.

The OBD reasoner can reason from annotations at the lower levels of the taxonomy to the higher levels. Given that Danio rerio exhibits a phenotype P, the OBD reasoner infers that Danio exhibits the same phenotype P. This is reasoning up the taxonomy, using the subsumption relationship between Danio rerio and Danio. This is possible because the annotations to each taxon are (implicitly) existentially quantified. The annotation Danio rerio exhibits uroneural is shown in (1). The semantics are in (2).

<javascript> TTO:1001979 PHENOSCAPE:exhibits PATO:0000467^OBO_REL:inheres_in(TAO:0000602) -- (1) </javascript> <math>\exists</math> X : instance_of(X, TTO:1001979) <math>\and</math> PHENOSCAPE:exhibits(X, PATO:0000467^OBO_REL:inheres_in(TAO:0000602)) -- (2)

Given that Danio rerio (TTO:1001979) is subsumed by the genus Danio (TTO:101040) in the Teleost Taxonomy as shown in (3), it is possible to infer that Danio exhibits uroneural (4).

<javascript> TTO:1001979 OBO_REL:is_a TTO:101040 -- (3) <TTO:101040 PHENOSCAPE:exhibits PATO:0000467^OBO_REL:inheres_in(TAO:0000602) -- (4) </javascript>

Inferring down the taxonomy, that is using assertions at higher levels to extract inferences at lower levels, requires universal quantification. For example, the assertion that all Siluriformes do not exhibit basihyal cartilage can be captured using OBD semantics as shown in (5). The universal semantics of this assertion is shown in (6). Siluriformes directly subsumes Ictaluridae as shown in (7). From (5) and (7), it is straightforward to infer that Ictaluridae lack basihyal cartilage as shown in (8).

<javascript> TTO:1380 PHENOSCAPE:exhibits PATO:0000462^OBO_REL:inheres_in(TAO:0001518) -- (5) </javascript> <math>\forall</math> X : instance_of(X, TTO:1380) <math>\and</math> PHENOSCAPE:exhibits(X, PATO:0000462^OBO_REL:inheres_in(TAO:0001510)) -- (6) <javascript> TTO:10930 OBO_REL:is_a TTO:1380 -- (7) TTO:10930 PHENOSCAPE:exhibits PATO:0000462^OBO_REL:inheres_in(TAO:0001518) -- (8) </javascript>

The problem with using top-down inferences using universally quantified statements is that currently there is no way to distinguish these from existentially quantified statements. We use the PHENOSCAPE:exhibits relation for existentially quantified statements. Using the same relation for universally quantified statements would make it possible to extract incorrect inferences given the current configuration. Consider the subsumption relationship between Danio and Danio choprai shown in (9). If there is no distinction between existentially and universally quantified statements, it is possible to infer from (9) and (4) the erroneous conclusion that Danio choprai exhibits uroneural (10). At present, there are no annotations to Danio choprai.

<javascript> TTO:1052801 OBO_REL:is_a TTO:101040 -- (9) TTO:1052801 PHENOSCAPE:exhibits PATO:0000462^OBO_REL:inheres_in(TAO:0000602) -- (10) </javascript>

Recall that the reasoner works in sweeps. It extracts one set of inferences (Inf-1) from the assertions (A) in its first sweep. In the next sweep, the reasoner pulls out a different set of inferences (Inf-2) from the assertions A AS WELL AS the inferences Inf-1 from the previous sweep. The reasoner repeats these sweeps until no new inferences are added. This is why the reasoner will likely infer all taxa exhibit all phenotypes if it is used to reason both up and down the taxonomy without checking for universal and existential semantics.

Possible solutions

In this section, we discuss possible approaches to resolving this issue with reasoning both up and down the taxonomy.

=Different relations for different purposes

In classical first-order logic (FOL), all relations and properties asserted upon concepts (or taxa in the case of Phenoscape) are inherited by the subsumed concepts. This is because by default, all assertions about the concepts are universally quantified, i.e. hold true for ALL instances of the concept. If all cars have four wheels, and if all SUVs are cars, then all SUVs have four wheels. This is the way of top-down, classical FOL inferencing.

In Phenoscape, we have adopted the OBD schema of modeling concepts, wherein all assertions to the concepts are existentially quantified, i.e. the assertion is true with at least one instance of the concept. This is very convenient for the life sciences, where exceptions are so prevalent. As a ready example, consider how the duck-billed platypus easily overrules the "all mammals are viviparous" rule. Further, existential quantification allows us to reason up the taxonomy. If some Teleostei exhibit round fins, and all Teleostei are Ostariophysi, then some Ostariophysi exhibit round fins.

By default, we use the PHENOSCAPE:exhibits relation to link taxa to phenotypes using existential semantics. Using the same relation to model universally quantified relationships between taxa and phenotypes, would cause incorrect inferencing and loss of data integrity. The easiest way to address this issue is to use different relations; one for universally quantified relations and the other for existentially quantified relations. Let us call these relations PHENOSCAPE:all_exhibit and PHENOSCAPE:some_exhibit respectively.

Now the OBD reasoner uses the following rule to extract inferences up the taxonomy using the PHENOSCAPE:exhibits relation.