Proposal for distinguishing traits and phenotypes

[dosumis]

• All Trait and phenotype patterns have a modifier. This may be phenotype, trait or children of these terms.
• Modifier: normal/abnormal has parent class phenotype (?) This replaces deviation(from normal)
• Additional modifier - trait

=> two hierarchies: Trait & Phenotype

Draft PATO structure proposal

• modifier
  • trait
  • phenotype # formerly deviation(from normal)
    • normal # This is here already
    • abnormal
    • ...

GCI: has_modifier some trait disjointWith has_modifier some phenotype

Metamodelling: Trait must only be used as a modifier of E + Attribute. PATO terms that are attributes (e.g. color) , not values (e.g. red) will be indicated via an AP. This is necessary because attributes are parent classes of values in PATO.

[dosumis]

phenotypic qualifier . normal # formerly deviation(from normal) . abnormal . . pathological
trait qualifier

defined classes:

trait EquivalentTo: has_modifier some 'trait qualifier' phenotype EquivalentTo: has_modifier some 'trait qualifier' . normal phenotype EquivalentTo: has_modifier some 'normal' . abnormal phenotype EquivalentTo: has_modifier some 'abnormal'

[cmungall]

What's the rationale for this? vs having this be a metaclass (e.g obo subset) distinction?

Note this is different from what some trait ontologies (OBA, TO) do, cc @marieALaporte (but OBA and TO already differ in that they do not use subq, as they are not is-a overloaded)

[matentzn]

@cmungall You prefer traits to simply be superclasses of phenotypes? For me it could go both ways, but I would appreciate the phenotypic qualifier category in the ontology to narrow down the range of the has_modifier relation in the patterns and make sure people are not using 'abnormal' as a quality.

[marieALaporte]

In the TO, we define a trait as being what is measured (Entity + Attribute) and phenotype as the value of a given trait. e.g. leaf color is a trait and green leaf is the phenotype. Traits and phenotypes are in the same hierarchy. We don't usually use normal/abnormal as a red leaf can be normal in some plant species, but can also be the result of some stress on the plant. Normal/Abnormal can be inferred depending on the experiment and trait measured. I understand that this is probably different in humans.

Regarding the patterns, we have 2 different pattern files, one for the regular EQ traits and one for the phenotypes. The patterns are in fact the same so I can get why @matentzn would like to be able to put restrictions on the Q part in order to avoid mistakes. But if I take the example of long or short or absent, I guess that they can be classified as being both normal and abnormal depending on the trait, the species, ... So not sure how you want to do it but that may be because we don't have the same definition for trait and phenotype

[matentzn]

@marieALaporte such a pity you are not there on Sunday! This would make for a good discussion. Can you point me to your current list of patterns? Are they here? Which ones are the ones you are referring to here?

That a trait is 'what is being measured' (Entity + Attribute) I think is what we think as well. Lets see what @dosumis says.

[marieALaporte]

Yes, that's the list of the patterns that we have in the TO.

I wish I could have been there too! I have been traveling so much lately that I also missed almost all the Thursday meetings... but things are calming down so I should be able to participate to the next ones :-)

[dosumis]

That a trait is 'what is being measured' (Entity + Attribute) I think is what we think as well. Lets see what @dosumis says.

Yep. The ticket is incompletely specified.

Current situation: PATO has attributes (colour, size) as superclasses of values (green, increased size). This means that traits (entity + attribute e.g. leaf colour, head size) are superclasses of phenotypes (entity + value, e.g. green leaf). We could trivially make the trait heirarchy into sibling classes by using a qualifier/modifier for traits - as we already do for abnormal phenotypes.

I'm agnostic about whether we should do this, but various members of the phenotype-ontologies alignment working group are uncomfortable with traits as superclasses. For this reason, I'd like to discuss this issue in the workshop - presenting this option as a practical way to achieve parallel heirarchies without churn in PATO - and leaving open the option of grouping logically. The implementation in trait ontologies could potentially be in root classes, avoiding churn in these ontologies too.

This ticket also proposes a similar solution for distinguising phenotypes, normal phenotypes & abnormal phenotypes which I hope will be less controversial.

[balhoff]

It's very convenient for traits to be superclasses of phenotypes, for semantic similarity. It would be helpful to hear more about what the downsides are for traits as superclasses.