Existing phenotype ontologies are arranged as simple classification hierarchies, with broader phenotype terms subsuming more specific ones. One notable feature of this design is that the classification of the phenotype is conflated with causal information connecting phenotypes. For example, the abnormal development of a thyroid gland is classified as a type of thryoid gland morphology. This is not a problem when using a phenotype ontology for search or simple grouping, but can pose problems when we try and use automated techniques to build phenotype ontologies or connect genes to phenotypes.
For example, as seen in this subset of the Mammalian Phenotype (MP) ontology:
is_a MP:0005379 ! endocrine/exocrine gland phenotype
is_a MP:0002163 ! abnormal gland morphology
is_a MP:0000681 ! abnormal thyroid gland morphology
is_a MP:0002951 ! small thyroid gland
is_a MP:0003421 ! abnormal thyroid gland development ***
is_a MP:0004663 ! athyroidism
is_a MP:0004696 ! abnormal thyroid follicle morphology
is_a MP:0004698 ! abnormal thyroid parafollicular C-cell morphology
is_a MP:0005355 ! enlarged thyroid gland
is_a MP:0013162 ! abnormal thyroid gland isthmus morphology
is_a MP:0013233 ! ectopic thyroid gland
The highlighted term is a developmental (process) term alongside and under a morphology term. The relationship between the development of the gland and its morphology is one of causation, rather than subtype.
For this subset of the Human Phenotype (HP) ontology:
is_a HP:0001939 ! Abnormality of metabolism/homeostasis
is_a HP:0010932 ! Abnormality of nucleobase metabolism
is_a HP:0004352 ! Abnormality of purine metabolism
is_a HP:0004368 ! Increased purine levels ***
is_a HP:0004369 ! Decreased purine levels ***
is_a HP:0010933 ! Abnormality of xanthine metabolism
is_a HP:0011814 ! Increased urinary hypoxanthine
The highlighted terms are amount (object) terms alongside and under a process term. The relationship between purine levels and purine metabolism is one of causation, with abnormalities in metabolism resulting in abnormalities in levels.
Furthermore, this structure doesn't give us an easy means of answering the question as to the effects of mutating genes known to be involved in the GO biological process 'purine nucleotide biosynthesis' (loss of this gene may lead to decreased levels, a gain of function mutation may lead to increased levels), or 'purine nucleotide catabolism' (having the opposite effect).
Furthermore, it can be argued that the generic structure of placing "X level" phenotypes under "X metabolism" is incorrect, because changes in the levels of a substance are not necessarily due to changes in metabolism - the changes may be due to abnormalities in import of export of the substance.
We have created a phenotype ontology CPN (Causal Phenotype Network) that separates the classification of a phenotype from its causal effect. Specifically, phenotypes that are due to processual changes form a disjoint hierarchy from those that are due to changes in attributes of physical objects such as chemical concentrations, cellular and organ morphology. The ontology is constructed by an automatic process using knowledge encoded in ontologies such as GO and UBERON.
CPN is species-neutral, and is intended to be used as a module by existing phenotype and trait ontologies. It can easily slot into existing OWL-based phenotype development workflows; whilst CPN itself forms disjoint hierarchies, end-product ontologies automatically collapse these distinctions into a simple and familiar subsumption hierarchy.
CPN is also extensible to creating sophisticated network models of disease...
CPO
The construction process makes use of knowledge encoded as logical axioms in existing ontologies.
For example, go-plus includes knowledge of how processes related to anatomical structures, and includes an inter-ontology axiom
'thyroid gland morphogenesis' SubClassOf results_in_morphogenesis_of some 'thyroid gland'
We treat this as a shortcut relation that can be expanded to:
'thyroid gland morphogenesis' SubClassOf regulates some (morphology and inheres_in some 'thyroid gland')
We define a pattern that finds all axioms of this form and generates two phenotype terms (one processual, one object quality) and additionally creates a causation relation between them:
Class: 'thyroid gland morphology'
EquivalentTo: morphology and inheres_in some 'thyroid gland'
Class: 'thyroid gland morphogenesis phenotype'
EquivalentTo: quality and inheres_in some 'thyroid gland morphogenesis'
SubClassOf: causes some 'thyroid gland morphology'
Whilst the two classes seem like close and even indistinguishable concepts, it is useful to draw a distinction, because changes in morphology may be due to other facts that changes in morphogenesis.
We can also apply the same pattern to developmental absence: if the development of a structure does not exist, then the structure can not exist (but non-existence of the structure could be due to other reasons).
When applying the same strategy to developmental processes such as growth, we can introduce directional qualities. For example, if a growth process is reduced in rate or efficiency, then the resulting structure will be (all other things equal) smaller.
The patterns also work on a cellular and biochemical level. For example: if the rate of disassembly of X is decreased, then this will increase the levels of X.
- "X assembly" phenotype -> X levels
- increased "X assembly" -> increased X levels
- decreased "X assembly" -> decreased X levels
- "X disassembly" phenotype -> X levels
- increased "X assembly" -> decreased X levels
- decreased "X assembly" -> increased X levels
The following GO biological processes can be used to create causal links as follows:
- "X morphogenesis" ==> X morphology
- "X growth" ==> X size
- ...
We note that currently in MP "abnormal X development" is classified under "abnormal X morphology". Note that there may be developmental abnormalities that leave morphology unaffected. There is a case for interpreting the MP class as actually being "abnormal X morphogenesis". TBD
- "X biosynthesis" ==> X amount (positive correlation)
- "X catabolism" ==> X amount (negative correlation)
- "X transport" ==> X amount
We can also use the GO axioms for the start and end location of a process to infer causation for location-specific amounts; e.g. levels of a substance in the blood.
Automated reasoning over CPN yields subsumption hierarchies that are heavily stratified - for example, process phenotypes will classify separate from object phenotypes. Morphology hierarchies will classify separately from volume hierarchies and so on.
For end-product phenotype ontologies, the current paradigm is to provide mixed hierarchies in which conceptually close terms group together. In order to support this conflation, OWL definitions for existing phenotype ontologies include a 'bundling' property (which we call 'has' here for simplicit). Without this property contradictions arise in reasoning.
In order to recapitulate existing conflationary structures, we include a property chain:
has o causes -> cause
This has the effect of classifying
TODO: abnormal
TODO: analyze linking of FYPO with mammalian ontologies
Of of the goals of this method is to effectively unify gene sets. Currently it is difficult to leverage gene sets deriving from phenotype curation in GO annotation, and vice versa. This framework provides a method for inferring the phenotype annotation for a GO annotation.
For example, a gene that is known to be involved in 'purine catabolism' (through some experimental assay) can be propagated to the corresponding phenotype term; e.g. we can infer that mutations in the gene can lead to 'abnormal purine catabolism' (of course, this may not always be the case, e.g. if the gene is rescued by a functionally similar or identical gene in the genome). Furthermore, we can infer that mutations in the gene may also lead to 'abnormal purine levels'.
See phenotype-ontologies/src/ontology/cpn/
id: MP:0013466
name: keratoconjunctivitis sicca
def: "inflammation of the cornea and conjuctiva caused by eye dryness which, in turn, is caused by either decreased tear production or increased tear film evaporation" [MGI:Anna]
IC <- ED <- {-TP, +TFP}
(TODO find better example)
abn swim bladder dev -> abn swim bladder -> abn regulation of buoyancy-> abn buoyancy
aldosterone
myocardial infarction