Workshop notes 2007 july 11

Artefact_object discussion

Frank Gibson joins by phone

Problem: Consensus to avoid creating classes unless there was a need for them. Having less subclassing is helpful for annotation use case. Artefact object is not helping and there is overlap with biomaterial in the current definition. We want to build a useful ontology not simply a correct one

Lengthy discussion:

BS:Need a place for instruments (Possible Synonyms for instrument: Device, apparatus, equipment) AR:under object BS:where cardinal part of? AR:under object CS:we eliminated device BS:I want to avoid biomaterial having siblings at a lower level AR:I want to get rid of biomaterial and push the kids up RS:I don't know why instrument and biomaterial entity are at different levels CS:lets bring Frank up to speed. We went through instrument and id classes that were not useful. Having less subclassing is helpful for annotation use case. Artefact object is not helping and there is overlap with biomaterial in the current definition. Unless we found a driving need for a class that is organising children below, this affected artefact, device and perhaps biomaterial entity. AR:Also reasoning, want there to be inferences for assertion for subclasses and if editing a paper less is more, if remove something and it's just as good, then you didn't need it FG:I wanted to make this disjoint with biomaterial, wasn't a device or an instrument, were thinking about HP:we can have cases of biomaterial that are synthetic FG:problem with the definition of biomaterial AR:agree We check the definition of biomaterial_entity BP:We talked about this definition during the call, hard to get a good definition, it's not perfect JF:we have a lot of things under biomaterial class, my understanding is that we want to refer to some of the entities but not as a biomaterial entity in all cases AR:what's a biomaterial entity, what's the use case for that JF:in mage was a sample for microarray HP:we can make our own definitions here RS:could be replaced with role. CS:in several other branches we don't have a clear measurement of separating them, we were going to let these sort themselves out, would be apparent which belong together based on definitions. FG:Issue is that dealing with mult branch under object, if 2-3 branches are merging that won't work AR:Chris you had an idea of what's natural vs unnatural, move all non natural things under artefact object FG:Of biological origin, things that occur in nature. Or could refer as a biological substance. CC:non biologic vs biologic and natural vs unnatural cut across LF:if we get rid of biomaterial and artefact we need to keep these containers. We can reclassify these AR:I thought we said just making a list shorter is not a good reason to add a class FG:I suggest two ways - things that are natural/manmade, or what you mean by biology - and everything else isn't AR:If I had to vote would choose man made vs. natural CS:Thinking about branches - we have a biomaterial and instrument, have these as they have been useful organising principle. May want to continue to work in that way. FG:There are more objects than material and instrument BS:list them: FG:polacrylamide gel CS:is material that is not biomaterial RS:organism is a portion of biomaterial CS:why 'portion of' BS:bad to use mass terms, hence use of portion. e.g. bad to have tissue and tooth, blood, bone, convert mass nouns to count nouns by using portion_of AR:does everyone get the distinction between mass and count nouns. BS:instruments and portions of material, are there any other objects we need, ecpsystem is on the site side. Need a term to deal with lab stuff, labware should be portion of labware KC:consumable? FG:we considered that BS:why can't we consider a testtube as instrument GF:things that do measurements vs things that are used BS:equipment? FG that could cover anything AR:if equipment and portion of biomaterial do you need anything else. E .g. Liquid chemical soln is now 'portion of material'. BP:there are liquid solutions that are biomaterials, don't like this solution, Top leve would be portion of material in general, where would atom go BS:portion of material. OBI should avoid it's own physics, atomics, tax of organisms. Should concentrate what happens in experiments BP:in lab, have mass spec that uses atoms, we need this BP:then why is equipment is not portion of material BS:we define as such, equipment serves a purpose, material doesn't BP:liquid buffer - made for expt purposes AR:answer is have class of liquid non bio solution, liquid chemical soln is too broad. BS:single inheritance for quals says that dark red is a child of red not dark LF:the spectrum is continous, we decide not to go there RS:we want to have both instrument and biomaterial or everything else. Why do we want to describe instrument - if we determine in a way that it generates data C:define in terms of function AR:everything has a function ac to current definition BS:needs fixing BP:could have utilizes instrument, to produce outputs from inputs - doesn't apply to heart or to portion of material. Doesn't apply to heart. BP and BS agree CS:not clear what's driving the discussion, other than trying to come up with a way to split CS:current way treat objects is mess. Frank has a point we need something other than biomaterial. Shouldnt put a special case for gels, as we would get an explosion. Gels annotate to portion of material, cow parts to portion of biomaterial CS:we put gels and cows under portion of material and instrument under equipment FG;equipment is a role, it is a device BS:don't like equipment? could substitute device RS:distinction between natural and artefact - congenic mouse line - is hard to make difference Liquid chemical solution -> portion of biomaterial DNA chemical solution->p o material BS:def of device is that we make it for a purpose BP:if liquid chemical soln is a buffer - is is for a purpose CS:are we getting hung up on labels? Device equipment - we want instruments and labware type stuff, challenge is names RS:need a sibling of instrument - labware/consumable under device BS:what are the examples? RS:pipette, test tube, - child of device. Consumable not a good synonym, as includes material. BP:test tube and buffer are better e.g. AR:problem with growth medium - it's a mixture - some biomaterial plus some non biomater. Ob agg also a problem AR:if we have liquid chem solution as a class, and child is DNA and salt in soln, DNA might be biomaterial BS:leather strap has a function, is a device not a biomaterial AL:of you chop up leather straps, then is a biomaterial. Can sintrument be a role BS:no, 2 kinds of things used as instruments, e.g. using Alan as a door stop BP:if we create one branch fr physical chem properties of matter, then soln has cells in serum will cause probs BS:!BJOERN BEING TOO PHILOSOPHICAL NOW! - OBI is for annotating data (pots and kettles) AR:all devices are man made. If use a non man made thing as an instrument this is playing a role, not an is-a instrument AR:could you split all man made devices? BP:define what you have, can be bio or non bio. AR:buffer is manmade BP:can define also how made it e.g. cell culture is biomaterial/material? BS:we need a third axis, biological_unit - include cell, organism BP:cell culture? BS: Is an object agregate BS:all in OBI will be unit instrument, parts of the unit screws in instrument, aggregates into platform. Material and biomaterial are non units, they are a continuum RS:simpler e.g. a cell can be a portion of organ, or an ob aggregate in cell culture BS:cells and cell cultures worth persuing the idea that these are portions of biomaterial. Organism is a portion of material. CS:have we made an improvement AR:want to work through the real world problems BS:mixtures, we choose when we define portion of biomaterial. Some general dissent on this. RS:how define biomaterial? AR:something which was alive, or derived from such, or synthetically made to mimic life. BP:two views in room. make h'archy for both GF:if we wanted to bring frank in, we have made agreement that we need to make a change, maybe Frank might like to hang up BP:we will promise to have some class of device. FG:if you work out a biomaterial, rest will fall into place. If you need that word at all. LF:does instrument have anything to do with machine FG:try not to use word machine. RS:we want to split on what generates data, so put freezer in same category in test tube FG:instrument: observe, measurement, record BS:easy test if material - on material - we can get more or less of it. CAn't get more of this machine BP:Agree - upper level separation is about physical chemical basis then BS:need organism under object as well. AL:why can't we have something that produces data FG:devices observe, measure record, not nec produce data BP:we can do the portion of stuff, it's straightforward if do along single countable thing and portion of, generated by a manmade process. Works OK. If we plit cleanly there will be no mult inhertitance. Stick with that CC said. CS:starting to be convinced that we can use this dimension to split, when I do an expt I have sep for things I am studying and this that is better represented by the first one.

Test Case 1

portion_of_material
  portion_of_biomaterial
    mixtures_with_any_biomaterial_constituent
    organism
    person
device (all man made)
  leather_strap (it is made for a purpose)
  labware (consumable NOT synonym)
     test_tube
  instrument (all play role of Instrument ?)
     measuring_instrument

Problems:
  chemical solution (of dna versus salt)
  heart (has function)
  mixtures (some biomaterial plus material) growth medium
  congenic mouse line
  fluorescent antibody (device or material(Ã "easy"))
  object
  Instrument (the role) (the class) ! <----
  "The correct" cell culture is an object aggregate (BS)
  Organization
  Buffer
  "Machine" (nanotech)
  peptide, protein .... peptide solution, protein solution

If can say: "Give me more of it" => Material
  But then: what about person? 

countable_objects
 antibody
 cell
 test tube
 organism
 peptide
 instrument
 heart
 congenic_mouse
portion_of_material
 isotropic
  growth_medium
  buffer
  cell culture
   portion_of_cell_culture in suspension
 anisotropic
   portion_of_heart_tissue
   portion of cell culture

Problems:
instrument
crystal
liquid
gas
heart
portion of material
Chinese has no mass nouns.

Conclusions:

• We will revisit, we need a use case for testing this. How will we deal with device? Suggestion to resort all OWL file based on these two things off line. And decide how to measure.
• BP:We could deal with man made as a high level class and infer class membership at high level
• BS thinks that man-made object will not work based on experience based on man made object, doesn't want to mess up what we have on what's a bad idea. Agreement for that. Also doesn't want physical/chemical at top, as don't know an e.g, of non physical/chemical objects and therefore is redundant.

2. Inferring hierarchies worked example from plan branch

Discussion

GF: if we want an inferred hierarchy need classes that will serve as bins for these inferred classes. AR:these are the defined classes, all named, given a def, and inserted correctly. Example: placebo, case control, randomized_trial. If drug plays role of placebo, infer placebo study. Classes added in test owl file to infer placebo_study, and case_control. PRS:Make it clear that this is not consistent with examples. CS:Is a new approach for classifying and doing definitions might help. Add OWL file from Alan demonstrating inference for studies. Inheritance is only inferred. PRS:from classification we have now I want rules to enter axioms, and how far to decompose. Is there a method to identify rules. Otherwise too empirical JF:empirical, but is a good thing, this will save us time. LF:have a lot of qualifiers, some will be common to PATO PRS:I asked if they were qualifiers, as they don't belong to PATO stay in OBI GF:mixing 2 things, this combinatorial thing of terms is mixing in something else pre vs post coordination. You need to build an application in concert with an ontology, classification will not work in this case from compound terms. GF:Alan showed can build a h'archy, not build from atomic components BP:GF is saying that we have to know what we want, we need to know what we need to create expressions. GF:I heard post coordination yesterday, this has a different meaning PFS:I want to do classification DS:There are ontology design patterns e.g. normalization. The example provided was a kind of design pattern. Refs:

• [[1]]
• "Untangling Taxonomies and Relationships: personal and Practical Problems in Loosely Coupled Development of Large Ontologies", Alan L. Rector, Chris Wroe, Jeremy Rogers and Angus Roberts. K-CAP 2001
• "Modularisation of Domain Ontologies Implemented in Description Logics and related formalisms including OWL", Alan L. Rector. K-CAP 2003
• [[2]]

• image to be added.

3. Use of inter branch relations esp. between protocol/protocol application, clarification of scope

Problem:plan vs. protocol application have duplicate terms and we need to resolve this. BS:Recommend PA should be the richest part of this, and protocol simpler. BP:Many PA terms were submitted that are not process, but have intent, get examples. e.g cell_sorting - use cells later, in some cases the cell_sorting is to generate data that is a measurement, process is similar, intent different. If we define inputs, outputs, utilizes we are talking at the plan level BS:can't see a problem, output of protocol appl can be either AR:can only define inputs in terms of plan not PA CS:useful to capture in OWL file and adding restrictions BS:OBI needs to choose, be an ontology of normal successfl biomedical investigations, or include the broken ones where it PA stops, etc. Then we can't formulate many axioms. We cna have protocol is realized by PA. Can't have has_output Data in this case as might stop where something fails. May subclass into successful vs. failed, and get many axioms for the successful. Or have a qualifier that 'usually yields output x). We decide to deal with the cases which are 'completed' and presumably successful CC:Say a broken protocol application is not a protocolapplication. ProtocolAppl is therefor e defined as completed execuation of the plan. All have some output.

Two properties added to OWL file 'is_realized_as' 'realized_as' Problem in OWL file plan is not a realizable_entity any more, and it needs to be BS: dependent continuant should be divided into qualities and realizables One of action items from last week, was to divide dependent continuants, into two types, was a step forward, but now we have problems. BS:Plans that are published, to realize them, you need to read and then plan is spec. dep continuant instantiated in the brain. Realizables are all under specifically dep. cont. No kids on BFO as likely children (information objects) are outside realm of BFO, no people in BFO.

AA:

BP:There are some other problems that need resolving e.g. digital entity.

We need to plan for updating to BFO 1.1, Barry assumed that we had. Added generically dep. cont but are no 'realizable gen. dep. cont'

The following changes have been made by Alan need cross checking vs these notes:

Citations: Roman Ingarden: 'Literary work of art' citation for relation. "  : 'Borderline cases of the Literary work of art citation for plan </b>

ProtocolAppl is the realization of an instantiated_protocol - an instantiated protocol is concretization_of protocol where concretization has a specific individual as the bearer, but generic protocol doesn't have a bearer.

Axiom: 'every PA involves a concretized_protocol, every concretized_protocol has a bearer possesor of some role describing the person that concretized it.'

'every concretized_protocol has_input, has_output'

BS: There are plans which are published (e.g. pdf). But, to realize such a plan someone needs to read it. Then you get a plan which is a specifically dependent continuant which is instatiated in the brain of the person running it. So, you can't realized a generically dependent plan without you yourself being engaged in a specifically dependent plan. BP: Even though there are bot GD and SD plan types, we shouldn't need to bother *showing* the GD one. BS: PA is the realization of an concretization_protocol which is the XXX of Protocol. This allows us to say that this PA is an instatiation of Nature Protocol ABC, but we also want to say that this same PA is an instatiation of a Protocol made by Jack that comes from Nature Protocol ABC. We need this three-tier thing as the GD protocol, by definition, does not have a specific individual as a bearer. Then we can have axioms like "every concretized_protocol hasBearer (some possessor of Role Experimenter)".

BS: Concretization is also to specific locales, etc. It gives it a specific context. Inputs/Outputs have to be specified in all three levels. BS:concretization is about context and person related gaps in the protocol.

BP:protocol for DNA purification, input is sample, otutput is DNA. We wanted to specify intended input and output. Generically dependent continuant has a specification for an input and output, doesn't have an input and output.

GF/AL:could just have an instantiation of protocol with input and output not protocol application, one or the other, do we need both?

AR:need in OWL need to be able to specify in a single place, and have it be propagated.

CS:would not be useful except at level of the ProtocolApplication only for my applications for annotation.

Some agreement that there is a need for these 3 levels of protocol, and that input and output are at the PA level

BP:Remembering problem, we were unhappy with intent in PA, now all PA is tied to plan and protocol and we can now then say what the roles are the level of the PA. Intent always comes now with plan/protocol.

AR:may need in more places than just PA

GF:protocol/pa has same types of things, problem with people needing annotation. Something will instantiate, this is the PA that you will instantiate.

AL:need to find the place where users are more comfortable.

BP:do we still have a problem. We need just to decide where it goes, and help people find it. We don't want to duplicate anything. Process will have the most detail, put it there.

Consensus: Inputs and outputs etc go on PA at the lowest of the 3 levels we have created, and don't make specific Plan or concretized plan until it is needed.

4. Use of roles, identify cases where move something from a current class to a role, provide a worked example of why for future use.

5. OBI only qualities - experimental only e.g. for qualities of instrument

BS:should be some OBI specific qualities if these are needed. We can't come up with good examples for instrument, are related to experimentally created biomaterials.

BS: OBI should have a small set of terms describing instrumental and other experimental qualities. Example qualities: randomized, fixed (cell), purified (cell), mounted (cell).

E.g. purified is a quality that is not in nature.

E.g. Study design qualifier: randomized

Consensus: we can have qualities in OBI. OBI would keep those qualities that don't exist in nature, as natural qualities will be stored in PATO.

AA:All branch editors to identify qualifiers and submit

AA:All relations in protocol application e.g. to be submitted to the relation ontology - Liju

6. Dealing with common object properties lot number, etc that apply to instrument/material

Stuff is different lot to lot, AR wants as a different type in this case, BP:thinks can be an object property. BP:Lot no doesn't deal with an ontology. Need a way to track instances of types outside OBI AR:Stuff is different enough worry is a different substance each time, not simply tracking BS:can we decribe the nature of the difference, or do we just think that lot 1 is different from lot2. GF:This is an individual, not a subtype, needs to be captured in a database. LF:could we use a data type annotation property, we agree, but doesn't address Alan's objection BS:there is no generalizable thing that you or we can say that can go into the ontology RS:batch, and buy aliquots BS:Alan is obviously wrong that these should be in the ontology as classes. BP:wanted to model as acquisition, by ab from vendor X, get a PA with it that also has a PA that assigns lot no. Process of acquiring is where the distinction is. JF:We put it in protocolapplication, you can give us the lot no if you want to. BS:you put it in the protocol not the PA. AR:type of substance doesn't depend on how you got it BS:lot no is not a type, disagree with you. GF:I get an idea but it's not clear what you want. It seemed for each lot you want a new class in the terminology. I could see creating a set of classes that describe lots, but the lots themselves are individuals. You are describing a class of one. AL:need something that says 'lot20' as a class, and portion of material would reference that. HP:suggest we take it to a vote. BP:we can have a purchased object as a class, purchased from, purchased by, then infer a class under foetal calf serum BS:purchased by is too granular. From OBI if people want to create children beneath 'lot' type then they can. AR:there is no class called lot. BS: Take leaf node "portion of freisian FCS". Then we say if user of obi wants to create further child classes that divide based on lot number, then they are free to do that. CS: Consensus is that we stop at a place where, if you want to use lot number, that's out of the scope of OBI. But as lot number is something that generally must be supplied, so have we failed our community by not providing a standardized terms? BS: We already have this problem with Affy machines 3.3, 3.4. CS: I buy Array U133, which also comes with a batch number which I also must provide. However, if we go with BS's idea, I don't have a term to properly describe this. BS: then we really *must* include these lots/batches as classes. RS: The distinction is arbitrary, as you could equally well make the class/individual distinction at aliquot, as sometimes people don't trust aliquots bought months apart. Where do we draw the line? AR: I don't like lot number, as it doesn't tell us much, and we can't figure out how to model it in a useful way. We need to decide if they're classes or not classes. In a class definition, I know what the defining properties are (e.g. lot number), which I don't know if they are instances. BS: In semantic web, if several instances share the same property, you can create a sharing class.

Summary of Discussion on whether lots belong in OBI or in OM, suggestion that Alan is biased as he's from semantic web and he's therefore overloading the ontology. If batch is an individual or a class, and what the consequences are for the ontology. There have been many suggestions, add a class for each lot (AR) use annotation properties (LF) exclude this info and put in a database (GF). Kevin provides some examples of tests that are used to qc lots. From pharma there is variability and need to be able to model that if important for your experiment. Lot is proxy_for some qc that has been done. AR:is_proxy_for wasn't meant for that, was a physical thing, proxy for a physical thing. GF: We want to have a lot no, we need a term for that thing, not nec. a relation.

BS:Proposal, we know we can have defined classes, can be on basis of something arbitrary. We tell users that they can create defined classes of lot membership. Treatment will be same as parent classes, definition is in the hands of the users. CS:How do we tell users of OBI? Where do we put it in OBI? BS:Docs will have that. BP:Haven't answered CS question, could build a class, acquired from AR:decide if classes or not classes. Can't see much argument that they are not classes HP:personal comment, that is not my view on what the consensus is in the room and also not my personal view.

AA.There was no resolution on what to include or how to model it, or even if it's in scope for OBI. Alan will write up an example and there will be more discussion. Joy.

Progress on OCI OCI will annotate a study, put on wiki, ask OBI to be authors when published. We have 800 terms from CDISC. How to send the others, we would place in OCI = OBI. Getting comments via email should so that collaboratively. Want some feedback. AR:agree each branch will triage some no of the terms. RS:we were proposing that they go to a branch and structure, we would do the branch and get feedback, more efficient BS:we get an application case in clinical case HP:do these terms have definitions? JF:yes, and tried to improve so pertain to OBI, but not completed, question was whether OCI did that or talked to branches. We will do ourselves and send to branches. BF:OCI usage, we will be part like any other community, acronym is to facilitate discussion, but no special status in OBI. BS: OCI will be part of OBI in the same way as any other branch, but the acronym has no "ontological legal status" (RS: a virtual acronym).

Question on if we need all reagents from all manufacturers in OBI - not yet resolved

7. Exploring granularity with worked examples e.g. ProtocolApplication

Chromium release assay [[3]] There could be an overarching protocolapplication, and there are clear sub protocols. There are some shared component. Same in biomaterial, you can always make things smaller dimishing returns. If leave intact then hard to classify these high level things. Do want to make them granular to do classification. Separation of material, pooling, combining, sep. materials, incubating etc. Material separation works well, pooling harder as often get a transformation. Not resolved in y'days discussion, progressed to:

%material_combination
 % admin_substance_to_cell_culture
 % admin_substance_in_vivo
 % material_pooling
 % sample_treatment_with_reagent
 % combination_of_sampling

Tried to make a series of components that are generic, partly suceeded.

Chromium release assay (CRA) measures killing of the cell Mixing two types of cells is common and can be part of many things. Desired outcome would be that want to see the choices and make some decisions. RS:material_treatment - e.g. with an interleukin, with another cell, would be part of the CRA. We mock up parts of protocol application, excerpted below from OWL file, all parents are not shown, some non local terms have been added for convenience.

• insert image*

GF:defining cell chromium labelling, wondering if should be defining the parent class, would we be using for annotation on admin a substance for a template on the child cell chromium labelling BP:good question, we have not decided. Could have higher level classes and people define their own HP:request that do the most common cases first.

%assay
   %chromium_release_assay (assay and has_part some cell_chromium_labeling, has_part some_cell co-co-culture, has_part separation_based_on_size some has_part some radioactivity_detection)
 %material_separation (has input min only 1, has output min 2)
   %separation_based_on_quality 
     %separation_based_on_size (input some material, output_some material) 
 %radioactivity_detection (has_input some_independent_continuant, has_output some data_object is_proxy_for some quality of radioactive *defining property)
 %material_portioning
 %material_combination 
 %administering_material_to_cell 
  %cell_chromium_labeling (has_input_cell, utilizes_some_reagent_chromium_51, 
 %combination_of_samples
   %cell_co-culture (has input min 2 and has input only cell culture has_output cell_culture, utilizes_instrument some incubator *also needs duration )
has_output_cell has quality some radioactive_cell) *defined class, necessary and sufficient
%roles
 %effector_role
 %chromium_release_assay_target_role
 %label_role
 %target_role
 %input_role
 %output_role
%biomaterial
 %cell
   %cytotoxic_T_cell (should be in the CTO)
   %EBV_transformed_cell (should be in the CTO)
   %medium
ChEBI_objects
   %triton_x100 (should be in ChEBI)
   %chromium_51 (should be in ChEBI) (has_quality radioactive, has_reagent role (should be confered by utilizes, to be checked in OWL file)
%instrument
 %gamma_counter
 %centrifuge
 %incubator
%labware
 %microtitre_plate
 %incubator
 %polystyrene_tubes   
 %paper_filter 
 %filtration_device (generates no data)
%OBI_quality (sibling of PATO quality)
 %radioactive (move to dependent.OWL)
 %!labelled (if we add this is saying same thing 2ways, is redundant. Role confers this!

note could have made cell labelling with has input some thing, has role label instead AA:Alan to add Quality branch to dependent branch AA:add duration to all process GF to explain to AR about assn of qualities in a class definition e.g. two out put materials have new properties BP:need a way to order e.g. has_successor Need to figure out a way to say an output from one into next

Notes by Helen Parkinson and Allyson Lister