Workshop OBI Vancouver 2008 Jan 30 notes

Wednesday January 30 Notes

• AI:move hypothesis, dt role, stat factor from role, add under new class specification suggest a new role - specifies? Plan/PA branch

• AI:processes can't bear roles, objective is used instead for e.g. negative control - this has been added to the action item wiki

• AI:BS will send out some notes on logical definitions for instance level relations to relations branch to prevent relation explosion.

• AI:communities to send terms to instrument branch to improve coverage. BB will send an OWL file for BIRN microscopy and imaging

• AI:Edit:suggested edit new class role assignment with group assignment underneath it if PA can be distinct from a Planned Process

• AI:Edit:Defintion of PA defined as being carried out to modify or generate data or materials in an investigation

• AI:discuss within Instrument branch consequences for making artefact object a defined class only to avoid problems where there are biomaterials that are also human generated.

OR can we could keep artefact object and add classes underneath it use this strategy only for biomaterials, as we think all instruments will be artefact objects. Would be weird to have instrument next to protein. MC will communicate to instrument branch. Biomaterial branch are going with the defined class and will use reasoner to place terms in h'archy e.g. under artifact object and some other parent

• AI:use roles for e.g. buffer in (bio)material rather than hierarchy - BioMaterial Branch

• AI: proposal how to define material: material is union of BFO object, BFP object aggregate, BFO fiat object part. Is now the root for biomaterial and artifact which are defined classes and will help OBI be multi granular

• AI:population now has material as a parent and not object aggregate as material is now a union of BFO things

• AI:biomaterial will attempt define these in vivo etc as a defined class and see if this works, these may be at assay level not material level, and these will likely be processes

• AI:DT can do the gating examples as a defined class into the ontology - branch action item

AI:Alan to write up an OBI import/mapping strategy doc as described at the workshop

AI:AR/JM/MC will work through a use case with the cell type ontology import needed for biomaterial and test this strategy and report on the results. This will be an OBI application ontology as soon as we start bringing into OBI

AI:develop a strategy for dealing with the OBO foundry ontologies and OBI application ontology if import works

AI:Add imports and all related issues as an agenda item for the next workshop SS

Barry Smith Role, Relation

• Time - OBI should deal with only biomedical investigations. Numbers, Time are outside OBI
• Information Ontology will be worked on by Alan and Barry
• Roles

 * if there is a red curtain in the theatre, the redness is not a role in this context
 * if a grid is split into 1km squares, km is not a role
 * need a word for reln between parameter, variables, and the investigation that uses these in PA - BS suggests SETTING as these are not roles like patient role

MC:you say that parameter, hypothesis, objective will live in Protocol. If you modify param during PA, is that a new PA?

BS:No, PA are changeable. If we set voltage outside param we need to ammend the protocol, maybe create a new one

BP:We always said that the we are looking at the end of the PA, and modify Protocol.

BS:In a protocol have a spec of a parameter, realize info entity, gets realized in a 2 stage process, when experimenter adjusts machine, and then when the machine runs with that setting.

BS:also state what will be measuring in the protocol

JF:what about the drug dosage

BS:same as a protocol setting

HP:what about the variables relating to the materials used in the protocol

PRS:this is still part of the protocol, similar to the inclusion criteria for a study. I think is part of the protocols

BS:may be part of protocols that are applied, not nec. specified in the chain.

HP:was thinking more about relation to experimental design, and how that is represented

JW:for me it's characterizing state of the input and the output

BS:agree, but inputs have choice

JW:where there are variable where these are not factors need to record these as well

RS:Think this deals with a lot of issues, analyte, factor, you are talking about kinds of factor now - settings, variables, when you realize these they become qualities of the instrument, the evaulant etc

AR:can be a process, can be a disposition

BB:includes serial no of the device, lot no of the reagent

BP:how implement? we have protocol, we need specification, and a relationship

BS:we need a partonomic analysis of protocols, some will be optional some will be necessary

AR:was hoping that we could use relations rather than rebuilt the h.archy

BS:if we make is specification, makes it easier as is realizable info entity

BP:this would be parallel to how we use role, tree for specifications

RS:now we need to deal with PA and investiagtion. Variables are spec at the level of investigation, as at the PA level dealing with analytes.

BS:this is an old issue. I think simple soln is that protocol doesn't specify all in planned processes, if you add all together gets specified.

BP:plan has an objective and specifies instructions.

BS:someone needs to create an anatomy of a protocol so we know which bits are optional

JW:where does this deal with the characterization of input and output

BP:new relationship that allows us to talk about specifications that do not come to reality

PRS:do we just need a relation specifies

AI:move hypothesis, dt role, stat factor from role, add under new class specification suggest a new role - specifies? Plan/PA branch

RS:we specify that we need to use organ liver - so we'll have an organ specification and will need to point to other ontologies

BS:yes, and for data format that would be also specified here.

RS:and that's where you point to other ontologies

BS:dealt with all cases where processes are bearing roles now, and resolved that

RS/BP:Sham surgery has a role

AR:bees dancing

BS:continuants have optional qualities, can lose parts, occurents have all their qualities and parts, can't lose or gain. If a process goes faster it's not acquired a new feature.

BP:bees dancing, communication is a planned process, objective is to send a message. Mock surgery is a planned process where the objective is to have a control

BP:This is in objective, we tie a process and the why in objective

BS:the process is not taking a role

AI:processes can't bear roles, objective is used instead for e.g. negative control - this has been added to the action item wiki

PRS:if objective is realizable entity then is a cardinal part of a protocol,

BS:as a specification

BP:all protocols have an objective.

BS:plans have constant and variable parts. Some things are left open. True of all parts and the whole plan. Objective will have some parts that are specific and some that are left open. The fact that plans are structured with constant and variable, we might specify coarse and fine objective e.g. cure for cancer, vs test a response to a drug in mouse liver. When we say objective, this is an info entity, the mult descriptions are possible.

PRS:want to be 100 sure. Objective is a planned process?

BS:objectives exist in plan and the realized objective. Need label these.

AR:objective, and achievement, or outcome?

BS:planned objective and realized objective?

PRS:sailing a boat, goal is to sail to India, I find USA. I don't realize the objective - what have I done

BP:you revise the planned objective

AR:Plan and Protocol now merged in to a single file

Roles:

* RO has class and instance level relationships. 
* OBI needs instance and instance_type relationships
* uniquely_identified is needed for the serial number instrument relationship
* issue with this relationship and URI as used by the semantic web needs to be resolved
* some instance level relations are primitive, some can be defined, not addressed in RO paper
* OBI needs to define it's instance level relns in terms of simpler instance level relns and this has not been done in logical statements.
* don't use names of related things in the relation name AR:thinks this is a consistency check and useful

AI:BS will send out some notes on logical definitions for instance level relations to relations branch to prevent relation explosion.

Melanie Courtot (for Daniel Schober), Instrument Branch

• OWL file up to date
• terms submitted are included
• finishing meta data clean up
• minor issues
• poor coverage, mostly flow cytometry
• issue with manufacturers and how relate to branch content
• expecting NMR, chromatography, NMR not in first draft
• current content=first working release
• need to discuss how import OWL files and what are the constraints
• consensus will be granular only to support specific use cases
• role and function confusion resolved y'day will be summarised
• labware -
• issues with how to represent by function
• how restrict use to instance level relations
• new relations requested
• commercial terms - where store?
• settings - now part of specification/PA

AI:communities to send terms to instrument branch to improve coverage. BB will send an OWL file for BIRN microscopy and imaging

BB:for BIRN we have microscopy and some imaging, we have all the meta data. If I submit as OWL would that be useful?

RB:We may not need to model in obi unless there is a query use case

RS:how would deal with a ruler? Or a pipette - a container

PRS:re consumables, another distinction, how deal with microarrays - used 1x. Is that an instrument or a reagent, where reagent is a role

MC:was thinking of using aggregate object so that composite material/instrument plays a role. Have role instrument, reagent etc

PRS:Instrument is a material with a role of instrument -

MC:under discussion also for biomaterial

AI:need a relation to connect PA to instrument and function and to work through how to represent that. LF, MC, BP

BB: how connect instrument to parts?

MC:using defined classes

BB:issues with part vs whole functions and inference

RS:parts of instruments can be parts, where is the line between instrument of part and instrument

MC:solved with functional parts, likely using these. Laser can be used as laser or as a part of.

RS:what part of an instrument has no function -

RB:the parts that no function in the experiment will not be modeled

Bjoern Peters, Protocol Application Branch

• specification etc needs adding from today's discussion
• worked on material transformation a lot (was biomaterial transformation)
• focus on text definitions, not much on OWL definitions as we need role and plan and they weren't ready
• meta data is good
• problem, induction of natural processes e.g. methylation, how acheived. objective is clear
• need a link between natural process to some other obo process
• have a conflict of labels for some classes
• conflict of labelling for e.g. cloning, overlapping terms in different ways - simply container classes now for non orthogonal terms
• Invitrogen schema, most is in PA, and needed to OBI-ize it, was a great schema took some resolution for PA
• material transformation complete
• assay incomplete
• substance detection - uses analyte - now problematic, here we'll have lots of things where can detect any quality, not exhaustive now. Need to add 'as the goal of'
• propose high level grouping of assays e.g. measurement of material quality, detection of label, etc. Issues with how separate DT from these assays, there is a dt but reporting on the process is natural, reporting as a dt is not natural. Proposal to group by what is the ultimate data output. Not clear yet if will work for all assays.
• planned processes need addressing : enrollment is needed, is not a data transformation, or a material transformation, we need the term, but would like to use planned process. We have things that are 'branchless' - should all go under planned processes. We could add more things here. Objective and specification will help
• group assignment and enrollment don't belong under the PA header. BS suggests remove PA as a subclass and move up h'archy. these are planned processes in which people are assigned roles.
• material transformation in PA or biomaterial transformation ? BP:sample or reagent transformations looks good

RS:when was biomaterial->material transformation.

BP:was a branch decision, Frank insisted

RS:do we need to link to a natural process? I can see why want to say methylation, but the natural process

BP:here because the labels conflict with processes outside obi like reverse transcription

RS:would call it generating making a molecule in a specific way. Suggest use as a chemical process

BB:and point to reverse transcribe

RS:you don't do DNA repair in a PA, you modify it. Not happy

BP:Ok these are as submitted

HP:labels are just labels, useful part in the definition

RS:induction of a natural process bad

BP:is a placeholder class, and we need to fix things here

BS:what would we substitute?

BP:back into higher level class, just a problem label

BS:deliberate methylation perhaps? or planned methylation

RS:lots of things mimicking natural process that are not under induction of natural process

BP:no GO term cell co culturing - maybe I need to change the label

RS:concern re h'archy

BP:there is no h'archy, we couldn't get to one, we focussed on defining things and later work on a h'archy, there is a bit of h'archy.

RS:if use an enzyme as a reagent you can talk about function, resist linking to processes

HP:suggest point at EC numbers rather than point at a GO process to avoid confusion

BS:worried about cloning, we have transformation_of in RO. Where have a thing that generates another thing, this is not transformation, is merging, or fusing

BP:we need to define were start with one or more, and end up with an output where multiple things might have happened

BS:cloning is not just material transformation as give rise to an organism

RS:merging and splitting are in material transformation

BS:then is simply a label issues between the class name and the RO name. That's OK but need to be aware of it and make the definition explicit

RS:many processes are composite processes and that's why things are hard

HP:but data is acquired that way. maybe need to use defined classes for common composite processes maybe even in a given domain

RS:care when say label - you don't detect the label, detect the photons, LIMS may care

BP:they can report specifically if needed. can use nesting for intermediate values e.g. measurements that contribute to the outcome

RB:we want OBI released, how far are you from functional PA

BP:in a better stage, is useable all the way for 2 use cases

RB:in next step for h'archy how long

BP:we have added general terms, and very specific terms for 2 assay. In next 4 weeks proof of prinicple for other use cases

RS:we have terms can use today. When you add to a h'archy will that affect us, we will still be OK. Things will just move

BP:h'archy is defined by what say, define an assay for e.g.

RS:I want PATO to define high level qualities does it do that?

BP:is unintuitive for me

RS:rate could be a general quality of a process

BP:h'achy could change if assay doesn't work in some cases

HP:seems to me that this will work, I can use assay as it stands I think

BP:text def says that there are mult steps in the assay

PRS:should that be defined here, or in the protocol?

BP:this is in the spirit of realizing a protocol, if you want to talk about steps, you can say all the parts, e.g. labelling is a material transformation

PRS:we have not finished PA and Protocol relation, BS told us that we need to determine the parts. I think we need to do that here

BP:The definition here is more generic, in terms of analyte, would change to another representation.

RS:where we talk about composite PA, even if there are sub parts material transformation and data transformation

BP:yes.All have some sort of dt, summing etc is a dt

PRS:perform a transcription assay, gene expression, I declare a protocol purpose is to detect mRNA<, material is a microarray, role consumable, utilizes some instruments.

BP:you just care about the assay? We don't have transcription assay yet. We need to add that Could define some assays.

RS:quality of rna in an evaluant, then that's a general class in OBI. When you do your application you are more precise about consumables and instruments when you instantiate

LF:NMR assay definition You are grouing assays on instrument

BP:not yet done, yes. We can classify as alt. h'archy. We need one assay h'archy and we went this way

RS:h'archy is based on types of measurements are made, ie the evaluant

BP:I would propose that, but you can define in other ways e.g. instrument manufacturers might want that

RS:I want to group all assays that measure virulence or function of a T cell

BP:also for our use case.

JF: do we need specific protocols to assign to a role, or a generic role assigment one?

AR:need both.

HP:could make a generic parent and then assign existing children

AR:relevant also to data processing, e.g. randomization of the data

AI:Edit:suggested edit new class role assignment with group assignment underneath it if PA can be distinct from a Planned Process

AI:Edit:Defintion of PA defined as being carried out to modify or generate data or materials in an investigation

AR:moves it up to a higher level, and that doesn't help

BP:I would like to move acquisition, up to planned process - things that come from outside a protocol application

HP:these are things that are not created by your protocol applications in your investigation, This is clear and I agree on planned process placement. Data collection is a bad name, confusing with the label with other parts of dt

Biomaterial Branch - Melanie Courtot

• OWL file is up to date
• term clean up in progress
• issue with def of biomaterial - new def added, material is BFO object
• BS suggest have material - as need biomaterial which has non bio parts
• need to allow to disambiguate biomaterials, artefact objects and instruments - use role and have biomaterial as a defined class
• action item for MC to take to instrument branch
• genetic info content is a problem - suggestion to send to sequence ontology
• issues with importing/mapping from outside ontologies
• what is a buffer, AR :a material with a role
• what is a mixture. we think population is aggregate
• artifact and natural AR:could use roles BP: also need for organism derived, or is derived from a planned process - mating can be a planned process for e.g. We will use artefact object instead. Need artfact object as a defined class for this to work, but we like the solution
• in vivo, in vitro, ex vivo etc BP:only true for cells, tissues in general things inside or outside the body hard for proteins. These are qualities of some biomaterials and need to restrict how is applied.

HP:do we lose artefact object?

AI:discuss within Instrument branch consequences for making artefact object a defined class only to avoid problems where there are biomaterials that are also human generated.

OR can we could keep artefact object and add classes underneath it use this strategy only for biomaterials, as we think all instruments will be artefact objects. Would be weird to have instrument next to protein. MC will communicate to instrument branch. Biomaterial branch are going with the defined class and will use reasoner to place terms in h'archy e.g. under artifact object and some other parent

AR:workaround, have an owl file, user interface promises that has the subclass axiom, so that these can be seen during development, but not for distribution.

AR:suggest that we only add mixtures that we care about, and not add general concepts

AI:use roles for e.g. buffer in (bio)material rather than hierarchy - BioMaterial Branch

AI: proposal how to define material: material is union of BFO object, BFP object aggregate, BFO fiat object part. Is now the root for biomaterial and artifact which are defined classes and will help OBI be multi granular

AI:population now has material as a parent and not object aggregate as material is now a union of BFO things

BP:e.g. manufactured protein, have proteins, these are materials determined by content, manufactured ones use the reasoner - classified under material and artifact object

BB:seems obvious that in vivo/in vitro etc need to be defined classes based on some properties

AI:biomaterial will attempt define these in vivo etc as a defined class and see if this works, these may be at assay level not material level, and these will likely be processes

BP:maybe attached to objects or processes as well as biomaterials.

RB:seems like we can test this

BS:these are locations of processes

BP:you'd define these as the environment of the cell or tissue then

AR:in vivo is to do with the participant

BP:cell in a liver measuring some metabolite, if it's ex vivo this experiment was ex vivo

HP:does this belong on assay, in vivo, ex vivo, etc.

BP:nothing else has used environment

James Malone - Data Transformation

• Cambridge Workshop summary, see wiki
• Done a branch review to look at complete terms corrected curation status
• 139, 51 complete - and h'archy OK
• we have some cases where we took a sensible view on suffixes
• chris S will do our external branch review
• use case docs next
• had to use roles to avoid multiple inheritance, will now use objective
• we used has_input and has_output - there are no logical definitions yet. these are future work
• people think we need to connection to algorithm and logical definitions of input and output

BB: is what's here partly the outcome of a microarray meeting

AR:dye swap merge is not mathematical, could logically define in terms of the data that is information entity

RS:this is merging replicates and is placed

BS:moving average -> moving average transformation or calculation

RS:still want to do gating

BP:there is a specific dt in e.g chromium release assay. for me natural definition would be in terms of the assays.

RS:we will run into the same issue with data transformations, and we can say what the inputs and outputs are without assay

AR:we can write owl statements for e.g. some of the terms. Some of these are composite, we can catch the commonalities.

BP:background correction is a process - and there is also an objective -

RS:issues with roles so we didn't do anything with some terms, and now the h'archy is flatter - we can go back to more h'archy

PRS:there are still some that are compound terms that include method and purpose e.g. false discovery rate. e.g. Box Cox is refers to the method from Box and Cox, most point to a function and get a base and need a parameter to get a function. If we go down the line of fixing qualification and a method.

RS:what is the difference? SOme things are well defined by the function and others are more complex. Things that were small and well described are based on function, in other cases we refer to a function

PRS:then we need to resolved the case where I need to an FFF and Tukey Bi weight correction, I need to find a property that I can set.

HP:there is no connection yet to algorithms anyway

BP:if you describe all steps along the way - people care about the outcome.

PRS:we need to provide the means to capture which algorithm, you just need a way to specify things.

BP:we just need to deal with that somehow

AR:inputs and outputs needs to be dealt with, or the objectives

RS:we can describe the output but not the objective

RS:runs through the gating example and dimensionality reduction

BP:this is not about matrices, this is about vectors, and projection of vectors

RS:OK

MC:move gate as a child of partitioning method

AR:no is a defined class in context

RS:we need a generic thing though like a vector partition

AR:you'll try to factor the neutral stuff, and make composite logical definitions

HP:if we have convincing use cases

BP:should avoid matrix terms.

RB:data were cells and vectors, and gating is a complex thing that uses terms everywhere

BB:if this terms to be practical send the maths things to Ivo and they might get on with it

AR:not the next thing that you do

RB:you can test gating context if you can define in terms of what is the input data

BP:some maths things will not have an objective, some that are tied to assays can be tied to an objective. for sine etc there should not be an objective, should be handled elsewhere. for an expt purpose then we should consider that

AI:DT can do the gating examples as a defined class into the ontology - branch action item

Environment Ontology Discussion - Susanna Sansone, Barry Smith

• GAZ - open source gazeteer, paris etc
• ontology - mountain, cave, bedding, all place types where organisms can live, should include the gut
• gaz - country names all place names that were ever used e.g. parts of countries that no longer exist, envo - reef
• Envo doesn't do natural qualities - PATO does that
• 1110 terms, in use by several organizations already

Ontology Imports - Alan Ruttenberg

• Alan outlines an import strategy

AI:Alan to write up an OBI import strategy doc as described at the workshop

JM:How does this look? Sep file

AR:yes, one external, one external derived, run a script to get parts of the info

JM:can see how could work, how it scales. Concern if we refer to a lot of other ontologies?

AR:could create externals by script if scaling issues

CK:could do with Alan Rector's methodology

AR:for some additional info e.g. pref term, may need to translate and will need to use the derived file fr that

HP:what's the next step - create an external file.

AI:AR/MC/JM will do a test with CTO into BioMaterial

CK:why do we need people curating it? Could do on fly

AR:not all URIs tell where a term is due to issues. Need to say is an instance of owl thing or a class. or may need to define domains and ranges so we might want to make assertions in OBI.

SS:we will create an owl file in all cases? We need to create cross product ontology and we can use OLS to do that. We would like OLS to serve this.

AR:this is a cross product ontology

SS:this we do when we need it. OBI will be pure, when you want for your application you import?

AR:no we will get things and they will live in OBI, and we will release that

BS:another use for importing, OBI to some community - you know they need some chem terms, you can do that

AR:yes you can package any way you want

SS:how will you have this packaged up for all

AR:there will be one file called externals and people can edit themselves

HP:I think we can build a default

RS:I thought a community would build their own, but that's not part of OBI

AR:what need, will be more critical what they get from an ontology. This is just a way to get external terms

RS:when we need an external term in OBI wasn't designed to build cross products

HP:concern about branding this into an OBI when we get all the other ontologies and saying this is OBI

AR:we could get a slim

RS:we should be getting a slim for an application ontology

SS:this is an obo foundry process

HP:this is an application ontology and we use it as a test, this is not OBI,

BB:like this, politic way as we can decide what we support. If we have obi and obi application, what would OBI be without these.