Converting Alzheimer’s disease map into a heavyweight ontology: a formal network to integrate data.
This is a modal window.
The media could not be loaded, either because the server or network failed or because the format is not supported.
Formal Metadata
| Title |
| |
| Title of Series | ||
| Number of Parts | 12 | |
| Author | ||
| Contributors | ||
| License | CC Attribution 3.0 Germany: You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor. | |
| Identifiers | 10.5446/38607 (DOI) | |
| Publisher | ||
| Release Date | ||
| Language |
Content Metadata
| Subject Area | ||
| Genre | ||
| Abstract |
| |
| Keywords |
00:00
DisintegrationComputer networkOntologyProcess modelingHypothesisMereologySelf-organizationOntologyXMLUML
00:16
Universal product codeFormal grammarComputer networkDisintegrationHill differential equationSocial classInstance (computer science)Constraint (mathematics)Level (video gaming)Domain nameAzimuthLecture/ConferenceMeeting/InterviewComputer animation
00:32
Complex (psychology)Programming paradigmDisintegrationPerspective (visual)Independent set (graph theory)ConsistencySystem programmingKnowledge representation and reasoningKolmogorov complexityMeasurementLevel (video gaming)SpeciesSign (mathematics)Standard deviationMechanism designScale (map)MultiplicationState of matterTexture mappingPhysical systemDifferent (Kate Ryan album)Complex (psychology)SpeciesDifferential geometryProteinInformationOpen setConsistencyFormal grammarSoftware frameworkOrder (biology)Perspective (visual)INTEGRALComputer fileMeasurementLevel (video gaming)Knowledge representation and reasoningOcean currentEvent horizonAffine spaceComputer programmingVotingWordDisplacement MappingSpacetimeArithmetic meanMetropolitan area networkFood energyDefault (computer science)AdditionComputer animation
03:46
AutomationTheoryFile formatNumerical taxonomyOntologyTime domainProcess modelingGraph (mathematics)DisintegrationTerm (mathematics)Knowledge representation and reasoningSystem programmingFunction (mathematics)outputEuclidean vectorKnowledge representation and reasoningPhysical systemContext awarenessProcess modelingDescription logicLebesgue integrationInformationMechanism designExpressionOntologyConnectivity (graph theory)Texture mappingGraph (mathematics)outputFunction (mathematics)Personal digital assistantGoodness of fitGEDCOMVapor barrierSpacetimeMetropolitan area networkComputer animation
06:25
Process modelingOntologyDisintegrationSocial classBuildingHierarchyCategory of beingDescription logicConstraint (mathematics)OntologyKey (cryptography)Social classAxiomatic systemProteinTerm (mathematics)InformationProcess modelingFunction (mathematics)Multiplication signCategory of beingProjective planeoutput1 (number)Physical systemBootingNeuroinformatikComputer programmingExpressionJunction (traffic)Computer animation
08:12
1 (number)Product (business)DisintegrationProcess modelingOntologyFunction (mathematics)Social classWechselseitige InformationSimulationProcess modelingFunction (mathematics)Complex (psychology)outputSocial classProjective planeMetropolitan area networkMathematicsCASE <Informatik>Computer animation
08:57
Data conversionProcess modelingInferenceHermite polynomialsDisintegrationCellular automatonMultiplication signOntologyWordGame theoryGroup actionProgram slicingSpacetimeComputer fileProcess modelingCellular automatonFile formatRule of inferencePlug-in (computing)Scripting languageComputer animation
09:57
Data conversionDisintegrationSpeciesSocial classoutputMappingSocial classAssociative propertyInformationProcess modelingElectronic mailing listCurvatureState of matterProteinFraction (mathematics)QuicksortLink (knot theory)Key (cryptography)Computer animation
10:40
InferenceDisintegrationTranslation (relic)Landau theoryAutomationInformationProteinProcess modelingoutputFunction (mathematics)Symbol tableTranslation (relic)Knowledge representation and reasoningOntologyMappingFrustrationConsistencyMathematical analysisLibrary (computing)Error messageSimilarity (geometry)Computer animation
12:04
InferenceMessage sequence chartDisintegrationMappingTexture mappingProcess modelingSocial classDirection (geometry)Key (cryptography)ResonatorMappingDifferent (Kate Ryan album)Translation (relic)Electronic mailing listComputer animation
13:02
Euclidean vectorProcess modelingData conversionDisintegrationSoftware frameworkKnowledge representation and reasoningoutputOntologyScale (map)MultiplicationMassSession Initiation ProtocolMaxima and minimaExecution unitMereologySet (mathematics)Process modelingMachine visionAssociative propertyConnectivity (graph theory)Direction (geometry)Fraction (mathematics)Right angleSound effectBit rateComputer animation
14:20
DisintegrationIndependent set (graph theory)Process modelingBit rateInstance (computer science)Social classLebesgue integrationCategory of beingGraph (mathematics)ConcentricTable (information)Event horizonComputer animation
15:05
DisintegrationTablet computerQuicksortForm (programming)Set (mathematics)ConcentricSocial classLatent heatInterior (topology)ProteinInstallation artProcess modelingSymbol tableMatrix (mathematics)Presentation of a groupPhysical systemCategory of beingWater vaporTexture mappingKnowledge representation and reasoningComputer animation
16:25
Independent set (graph theory)Process modelingOntologyKnowledge representation and reasoningScale (map)DisintegrationPerspective (visual)CASE <Informatik>Medical imagingInformationKnowledge representation and reasoningPerspective (visual)Scaling (geometry)Different (Kate Ryan album)OntologyFormal grammarProcess modelingLatent heatTexture mappingFunction (mathematics)ConsistencyLevel (video gaming)MappingExpressionACIDEvent horizonAffine spaceSpacetimePresentation of a groupPhase transitionSound effectSurjective functionRange (statistics)Computer animation
18:06
DisintegrationRankingMedical imagingComputer animationLecture/Conference
18:20
DisintegrationXMLUML
Transcript: English(auto-generated)
00:00
Hi everyone, I'm Vincent and thanks, I'd like to thank the organizers to give me the opportunity to present a part of my work. I'm working on the hypothesis that ontological upper models with strong constraints could manage lower level of knowledge as classes
00:23
and then data as instance. Currently, I'm working on the Alzheimer's disease domain. So, in short, Alzheimer's disease is one of the main neurodegenerative disorders
00:43
and despite high efforts on research, there is one dominant branding and no curative outcomes, mainly due to a complexity that is not totally understood by researchers.
01:02
OMIX technologies open new perspectives to a global analysis of such a disease but the integration of the heterogeneusity of this kind of data is still a challenge. In order to deal with these problems, some community developed systems
01:24
medicine-based disease maps that are consistent resource frameworks to provide a global representation of current curated knowledge and to integrate the complexity of a disease at the molecular or phenotypic levels
01:45
and so with different information on genes, RNAs, metabolites, it is adapted to the diversity of OMIX measurements. So, the ALSPICE-1 is one of these disease maps focused on Alzheimer's disease
02:06
and here it's all the molecular reactions involved and new as involved in Alzheimer's disease mechanistic and so it's representing more than 1,000 species.
02:25
So, species could be genes, RNA, proteins, protein complex, phosphorylated proteins. So, it could be the same species but at different states in the systems. There is also 100 phenotypes and all is integrated
02:43
in more than 1,000 biochemical reactions. This knowledge is important in quality because it is manually curated resources and currently it is stored in XML-like files such as SBGNs or SBML.
03:09
It's also a multiscale resources while you can design a pathway that involves several biochemical reactions
03:20
but there is a lack of formalism regarding consistencies of the described reactions of the naming of the different species and in fact pathways and chemical reactions are not formally linked
03:41
in this kind of resource for ALSPICE-1. So, we hypothesize that depending on the expressivity of the resource you can embed other ones. So, here we have XML-like resources
04:03
and you can have an increase of expressivity by RDF, AL and AL-GL and then ontologies and particularly strong ontologies with description logics could embed and add and contribute with automated reasoning
04:26
to automatically reclassify or re-categorize the initial information. Moreover, as we are working on systemic context
04:43
we wanted to keep the system representations because it's in fact a kind of graph. A system is a process that transforms inputs and outputs and could be mediated by other participants
05:00
so inputs, outputs and mediators are participants and so you can represent it in an ontology as a process as participants, only participants as input, at least one participant as output, at least one participant and as mediators, only participants.
05:24
Moreover, processes can be aggregated in higher processes and so a process could be also defined as a component process of other processes.
05:45
So, we wanted to work with simple models and see if we can integrate knowledge and data about Alzheimer's disease molecular mechanisms. So, we first designed an ontological upper model
06:05
called Alzheimer's Disease Map Ontology. Then we converted Al-Spiceway in all and managed it with logical reasoning and after we aimed to integrate omics data related to Alzheimer's disease.
06:25
So, to design our upper model ontology we were influenced by terms of the system biology ontologies and also by the vocabulary used in MEPN projects
06:44
which can be used to design the disease map. So, we take terms of interest in the first time. When it is derived from SBO we keep the information about the original URI
07:04
but as during ER key binding we have some modifications compared to SBO we prefer to keep information as a class annotation more than making real extractions
07:23
to avoid ambiguity with the previous works. And after building ER key we add some properties with strong constraints and axiomatic between classes. So, for example, a protein complexation
07:44
has output at least one multimers and has input, for minimal, two proteins. That could be single proteins or multimers
08:00
and we can also add some disjunctions between genes mRNAs and proteins to gain informalisms. So, the upper model consists of 140 linked classes
08:21
with 42 processes and 83 participants' classes and each process is formally defined depending on their participants. So, a transcription has input some genes
08:41
and has output some unprocessed mRNAs and other examples of protein-protein complexations as I said, has output some heteromers and has input, at minimal, two gene products. Then, in the second time we converted ALT-Spiceway in OWL format
09:05
So, ALT-Spiceway is an SBML file that we extract in CSV with cell designers. Then, using some scripts we build some multi-sheet Excel files
09:23
to be managed by protege plugins called SELFIE and we also have our upper models and then the ALT-Spiceway is automatically using some rules
09:41
integrated in a single ontology with the upper models. And then we manage it with automatic reasoning using the RMIT reasoners. So, to describe this state
10:02
following the map extractions we have a flat list of processes which are described for SFRP Wnt associations with its inputs and outputs and so participants are in the class participants
10:21
and could be proteins and we have subclass annotations with labels linked to UDProt if there are some those processes can be linked to PubMed information in fact we kept all the information contained initially in the map
10:40
and then if we look at the map there is some kind of inconsistent so here is a symbol for translations here is a symbol for RNA and here proteins and in the map you have some so it's manually designed map there is some mRNA
11:02
were translated in proteins but we also have some genes that are translated in mRNA which is logically inconsistent and if we build an ontology with the initial knowledge we have that this is the representation
11:21
of the two processes this process is translation and this process is also a translation and using the OPIR models where we designed that a translation has input only mRNA and as output
11:41
some proteins while transcription has input genes and as output mRNA you can automatically insert that this process is a translation but this one is not a translation so in capital letters is the original information but it is a transcription
12:02
and so following the reining of the automatic reasonings our previous flat list is categorized depending to the process we defined in the OPIR models and so we have here some transcription that are subclass of transcription
12:22
but initial translations are also corrected as transcriptions and in the original maps there is many reactions that are not labeled well, they are just all called transitions
12:43
and by defining the different granularity of process we can infer that these four processes describing the map are acetylations and we can do the same for different phosphorylations and so on for each process
13:00
we described and so we also wanted to formally link the pathways and the biochemical reactions so we use another part of our model that a process is component process of a process and so we can develop the model as reactions are component process
13:21
of pathway and if we defined a process that has participants some wind is equivalent to a component process of some wind signaling which is a pathway you can automatically infer
13:43
that the SFRP wind association process is a wind signaling process and so it narrates from a B component process of some wind signaling and as pathways are in fact artifactual human visions of a set of reactions
14:02
you can design it as you want in a flexible way so with this with the mathematical reasoning we can formally link the pathways and the reactions
14:20
and so to finish we aim to integrate biomedical data as instance of our different classes so usually biomedical data are in tables more or less structured and you can then
14:40
structure them in triples using an RDF graph and so data are linked to subject and data and also subject have some data properties such as score, rate, SNIP, concentration for data
15:00
or agendas for subject and so we have simple RDF models with some subjects so here S1 to S4 and some data concerning genes for genomics, mRNA for transcriptomics or protein for proteomics
15:21
but we can also imagine metabolites, concentrations for metabolites and if we use our models with process participants participants can be divided with a disjunct gene mRNA proteins we can instantiate
15:41
our class with the specific data of a gene for subject, mRNA for another subject and in fact here is an abstract representation of all the extracted maps in Alzheimer's disease
16:01
for Alzheimer's disease and you can also add or insert some new properties between protein, mRNA and genes and all in all your first set of data is integrated in the whole
16:21
system represented by classes so to conclude an ontological model with high expressivity may embed a disease map such as ALT-space-Y and by doing this we increase
16:40
the formalization and the satisfiability of the disease map avoiding some logical inconsistencies we can also formalize different scales of representations that open perspective to add some phenotypics information as output of pathways
17:00
and we can enrich interrelationships between biomedical data such as the heterogeneity of omic data and so it opens some perspectives the next step is to
17:21
integrate the Alzheimer's disease map ontology with the Alzheimer's disease ontology which exists but represents more clinical specification so we can increase the level of representation and integrate also imaging which is important in case of
17:42
neurodegenerative disorders and also one advantage of upper models is that it might be relevant for other disease maps while disease maps community is a large community and so we can
18:02
unify the representations of different diseases so to finish I wanted to thank you for your attention and warmly thank people who work with me and the different funders Thank you very much Thank you
18:21
Any questions?