From: Chris M. <cj...@be...> - 2010-03-10 22:51:04
|
On Mar 10, 2010, at 1:54 PM, Robinson, Peter wrote: > Chris, > > I think I am not quite managing to communicate clearly what I am > thinking. The remark about history of seizures is meant to say that > it is not necessary for the physician to actually be in the room > when the seizure occurs in order to make the diagnosis seizure. The > remark about modeling seizures was meant to say that a seizure is an > incredibly complex process affecting multiple centers in the brain > and the HPO term is not meant to be a physiological model. I think I > am speaking in a way that is immediately comprehensible to anyone > with medical training, and my purpose in putting perhaps incomplete > definitions on the table is to get reactions on things that seem > dodgy or incomplete. Apologies if I misunderstood or over-interpreted. However, I think all my comments still stand. Apologies also if I seemed over-critical. I appreciate the work you've put in to adding these definitions, and they help tremendously. But I really think we need to get at the root issue of what the instances of HPO terms are. I'd be interested in your thoughts on my specific points below. > In the minority of cases one is able to identify a physical > abnormality that gives rise to seizures, but for most patients all > one can do is speculate that there must be something abnormal > somewhere in one or more locations of the brain that has led to one > or more seizures. Therefore, I do not think that it is a good idea > to define the seizure hierarchy on the basis of physical > abnormalities. I do not recommend building the seizure hierarchy entirely on the basis of physical abnormality. There would be multiple axes of classification - location (e.g. temporal lobe), etiology, psychic manifestation, etc. Note when I say "seizure" here I mean seizure, i.e. the process. We don't need to worry about the epistemological aspects. If you don't know what the specific abnormality is, you don't say anything about it. > Note that a few percent of all people experience a single seizure at > some point in their life. > MDs do not call this epilepsy, but rather, the history of having had > several seizures is interpreted to mean that there is a disposition > to have (more) seizures. At this point I am choosing to use the term > "seizures" as a parent term, because there are some kinds of > recurrent seizure that are not epilepsy (e.g. febrile seizures in > children). However, there are genetic diseases characterized by the > fact that affected children have febrile seizures. This is all easily accommodated for in the scheme I present below > The structure of the ontology in this place will reflect current > medical norms in the classification of seizures. Any other is-a > links will be hopefully recreated from the PATO definitions. Let's see if we can work with this, stay within standard medical classifications and terminology and yet have an artifact that can integrate well with other obo ontologies (I'm not quite sure I understand the motivation, we already have MESH and so on to do what you want to do, but anyway) If I understand correctly, your constraint is that HPO must reflect medical norms and have classes labeled "Seizures" and "Epilepsy" and they must be arranged in a broader-narrower hierarchy like this: Seizures Epilepsy So how about HPO:Seizures = HPO:'abnormality of the CNS' and gives_rise_to PPO:seizure HPO:Epilepsy = HPO:Seizure and has_qualifier recurrent and chronic This sketch is somewhat simplified both clinically and ontologically. The relations used in the above could be expanded to something more specific that can be used in reasoning. We still need to explicitly represent the process (denoted here by the not-yet-born Pathological Process Ontology class PPO:seizure) Of course there is a massive terminological problem here. The solution is for the HPO class to have two labels. One would be the label used by the HPO tools you are developing ("Seizures", "Epilepsy"), the other would be labels used by the OBO Foundry (not sure exactly, but something like "prone to seizures". Perhaps "epilepsy" is fine). Does this sound along the right lines for you? I think we can work with this. Note we still need a place for the process. > I would suggest we work on the PATO definitions for these three > terms and see what design patterns work best. Then I will go with > that and rework the entire seizuire subontology. I think we are at the rework stage. > -Peter > > > > Dr. med. Peter N. Robinson, MSc. > Institut für Medizinische Genetik > Charité - Universitätsmedizin Berlin > Augustenburger Platz 1 > 13353 Berlin > Germany > +4930 450566042 > pet...@ch... > http://compbio.charite.de > http://www.human-phenotype-ontology.org > ________________________________________ > Von: Chris Mungall [cj...@be...] > Gesendet: Mittwoch, 10. März 2010 22:37 > An: Obo-human-phenotype > Cc: Maryann Martone > Betreff: [Obo-human-phenotype] seizures > > We are starting to tackle some of the more difficult classes in HPO to > assign logical definitions. In many of these cases we will require > some refactoring in the HPO, or at least clarification of the meaning > of some of the terms This is particularly true where processes are > concerned. We already discussed this to some extent previously on this > list in the case of "Syncope". > > Let's have a look at Seizures. > > HP has: > > .is_a HP:0000707 ! Neurological abnormality [DEF: "An abnormality of > the central or peripheral nervous system."] > ..is_a HP:0002011 ! Abnormality of the central nervous system [DEF: > "An abnormality of the `central nervous system` (FMA:55675)."] > ...is_a HP:0001250 ! Seizures [DEF: "Seizures are an intermittent > `abnormality of the central nervous system` (FMA:HP:0002011) due to a > sudden, excessive, disorderly discharge of cerebral neurons and > characterized clinically by some combination of disturbance of > sensation, loss of consciousness, impairment of psychic function, or > convulsive movements."] > ....is_a HP:0002197 ! Generalized seizures [DEF: "Recurrent > generalized `seizures` (HP:0001250), that is seizures that affect both > cerebral hemispheres from the start of the seizure, producing loss of > consciousness."] > .....is_a HP:0002121 ! Absence seizures *** [DEF: "Recurrent absence > seizures are `generalized seizures` (HP:0002197) that are > characterized by a sudden cessation of motor activity and by a blank > facial expression with flickering of the eyelids. There is no > convulsive muscular activity or loss of postural control."] > > I think this is getting much better. We have what looks like a clear > definition of `seizures` as an abnormality of the CNS that arises as a > result of an aberrant biological process (I thought it was the other > way round, ie the physical abnormality gives rise to the aberrant > process? Anyway, I'll trust the HP definition for now) > > Peter also notes parenthetically: > >> Note that the usage of "seizures" means that a patient has a history >> of muliple seizures, not that we are modeling the seizure process >> itself. > > This is a pretty major semantic shift, and at the very least deserves > mentioning in the definition itself. This qualifier raises some > questions about the nature of HPO - is it intended to be an ontology > that can be used for reasoning, and can integrate into the OBO > Foundry, or is it intended as a terminology with characteristics > geared towards one particular application. > > I would characterize the major distinction between a terminology and > an ontology as follows: with an ontology you know exactly what the > instances are. With a terminology you can afford a certain looseness > and use a term to mean either: a history; a process; or a some kind of > structural abnormality that is not necessarily giving rise to an > aberrant process. Humans are good at disambiguating, machines are not. > > If we are to treat HPO as an ontology, and use it for integrating data > with other sources - especially the multitude of neurological > resources that are using ontologies - then we must be clear about what > the instances are, otherwise we cannot integrate data. Note there is > no reason why HPO *must* be an ontology - it could be a terminology > with cross-references to a separate ontology or multiple ontoloies, > where the ontology is clear about these distinctions. The HPO could > still have a loose thesaurus-like hierarchy that is tuned for semantic > similarity searches, and the HPO could still be used as a bridge > between resources like OMIM and ontologies where reasoning is used. As > Colin said, "there's no shame in being a terminology". > > I will go on the assumption here that HPO is to be used as an > ontology. The first option is to treat Peter's remark as gospel, > overruling the current text definitions, and treat instances of HP: > 0001250 as being instances of patient histories. I think in this case > the class should probably be relabeled "history of seizures". The text > definition absolutely must change to reflect this too, as must the > logical definition. I don't think this solution is optimal though. > > I would instead recommend that general statements of the form: >> Note that the usage of "seizures" means that a patient has a history >> of muliple seizures > be accommodated in the use of the ontology, e.g. in annotations. > > i.e. rather than making the HPO class mean "history of seizures", > instead make the class refer to the thing itself, and make the > annotation mean has-a-history-of. I think this is vastly preferable, > and makes the HPO more flexible. For example, it would then be > possible to use the HPO term for a patient that has suffered a single > seizure rather than requiring an entire history. > > If we follow this recommendation we still have to decide what an > instance of HP:0001250 is. I would say there are two main choices > > * [P] the abnormal process or ensemble of processes > * [A] the structural abnormality of the CNS that gives rise to the > abnormal process(es) (unless treated etc) > > OGMS would say that there is a 3rd type of entity in the picture: > > * [D] the disposition inhering in the abnormal CNS [A] that is > realized as an abnormal process [P] > > I think this disposition complicates the picture too much and we can > ignore this for now. I believe entities of types P and A above are > sufficient. I think what I say follows is not incompatible with OGMS, > but at the same time requires a weaker commitment. We have the option > of additional types such as "history of Ps" later on, but as I said I > think this is better handled using P as the building block. > > In either case, my picture is undoubtedly too simple, in that there > are multiple Ps and As involved at different levels etc, but let's see > how far we can go with it. > > It seems to me that the label "seizure" denotes entities of type P, > whereas labels like "epilepsy" denotes an entity of type A. > > Unfortunately, HPO has this: > > .is_a HP:0001250 ! Seizures [DEF: "Seizures are an intermittent > `abnormality of the central nervous system` (FMA:HP:0002011) due to a > sudden, excessive, disorderly discharge of cerebral neurons and > characterized clinically by some combination of disturbance of > sensation, loss of consciousness, impairment of psychic function, or > convulsive movements."] > ..is_a HP:0001275 ! Epilepsy *** [DEF: "Epilepsy is used to describe > chronic, recurrent `seizures` (HP:0001250)."] > > This is surely wrong. HPO is saying that all instances of epilepsy are > instances of seizures. I suppose if we use the history-of > interpretation this could be justified, i.e. every instance of a > history of epilepsy is an instance of history of seizures (but what > about non-symptomatic e.g. drug-suppressed epileptics?). > > Again, there is absolutely nothing wrong with a terminology specifying > a fuzzy broader-than link between epilepsy and seizures, and using > this for the purposes of search, query expansion, semantic similarity > algorithms, etc. The HPO definition "epilepsy is used to describe..." > indicates that HPO really wants to be a terminology here. But if HPO > is to be an ontology (and in particular an OBO Foundry ontology > interoperating with multiple other ontologies) we need to be clearer > about the distinctions here. > > The NIF Dysfunction ontology actually inverts the relationship that's > in HPO! > > is_a NIF_Dysfunction:birnlex_12796 ! Nervous system disease > is_a NIF_Dysfunction:birnlex_12718 ! Epilepsy *** > is_a NIF_Dysfunction:birnlex_12748 ! Seizures *** > > This ontology is (I believe) mostly derived from MESH: > > Epilepsy *** [DEF: "A disorder characterized by recurrent episodes > of paroxysmal brain dysfunction due to a sudden, disorderly, and > excessive neuronal discharge. Epilepsy classification systems are > generally based upon: (1) clinical features of the seizure episodes > (e.g., motor seizure), (2) etiology (e.g., post-traumatic), (3) > anatomic site of seizure origin (e.g., frontal lobe seizure), (4) > tendency to spread to other structures in the brain, and (5) temporal > patterns (e.g., nocturnal epilepsy) (MeSH)."] > Seizures *** [DEF: "Clinical or subclinical disturbances of > cortical function due to a sudden, abnormal, excessive, and > disorganized discharge of brain cells. Clinical manifestations include > abnormal motor, sensory and psychic phenomena. Recurrent seizures are > usually referred to as EPILEPSY or "seizure disorder." (MeSH)."] > > Note also that NIF treats this as a disease, not a phenotype > > DO has had the good sense to obsolete DOID:2542 ! Seizures. > > Unfortunately, DO has this: > > is_a DOID:4256 ! rheumatism > is_a DOID:1045 ! Muscle, ligament and fascia disorder > is_a DOID:423 ! myopathy [DEF: "Myopathy is a peripheral > nervous system disease consisting of any abnormal condition or disease > of the muscular tissues; commonly designates a disorder involving > skeletal muscle."] > is_a DOID:440 ! Neuromuscular disease > is_a DOID:1826 ! Epilepsy > is_a DOID:11832 ! visual seizure *** > > Nuff said.. > > I have some simple recommendations for getting out of this quagmire. I > am convinced that following these recommendations will lead not only > to less duplication of effort across the OBO Foundry candidates, but > also less effort on an individual basis for the HPO developers, and > increased collaboration amongst the relevant experts. > > First of all, seed an ontology of abnormal processes. To make this > more tractable we could start with neurological processes - seizure, > syncope (as discussed previously on this list), and so on. This should > be a collaborative effort and involve domain experts (e.g. NIF for the > neurological processes). Ideally there would be something we could > reuse here, but I don't think there is. The first pass could be a very > simple is_a hierarchy, that unambiguously inherits from bfo:Process. > This will look in many ways like the GO biological process hierarchy > (and there may be some fuzziness in the boundary between 'abnormal' > and 'wild type'). > > So for example: > > abnormal biological process > abnormal neurological process > seizure > temporal lobe seizure > complex partial seizure > simple partial seizure > .... > > This gives us our core building blocks with which to construct classes > such as: > > * [A] chronic neurological disorder that gives rise of seizures > * [D] disposition that is realized as seizure (if you follow the OGMS > approach) > * [H] history of seizures (if there is really a need to pre-compose > this, rather than specifying this at annotation/usage time) > > We can make unambiguous fine-grained distinctions where required (e.g. > patient X experienced first seizure at age N), or very general vague > high level statements where required. > > We can further axiomatize the abnormal process ontology at our leisure > - adding part_ofs, N+S definitions, links to physiological models... > > Are there any objections to this approach, or should we go ahead and > get started? > > > > ------------------------------------------------------------------------------ > Download Intel® Parallel Studio Eval > Try the new software tools for yourself. Speed compiling, find bugs > proactively, and fine-tune applications for parallel performance. > See why Intel Parallel Studio got high marks during beta. > http://p.sf.net/sfu/intel-sw-dev > _______________________________________________ > Obo-human-phenotype mailing list > Obo...@li... > https://lists.sourceforge.net/lists/listinfo/obo-human-phenotype > |