OpenTox

Dr Nina Jeliazkova

Ideaconsult Ltd Sofia Bulgaria

Why integration framework for predictive

toxicology

March 17 2011

bull Challengesndash Chemical structures

bull Might be ambiguous

bull Might be error prone or time consuming to reproduce from publications

ndash Data bull Multiple formats

bull Implicit semantics often buried in human readable documentation only

ndash Modelsbull Tens of thousands available in software or in publications

bull Multiple software solutions mostly incompatible

bull Predictions reproducibility is time consuming and often hard to achieve

bull Automatic comparison of prediction results difficult

Why integration framework for predictive

toxicology

March 17 2011 Ideaconsult Ltd3

OpenTox Framework approach

4Ideaconsult LtdMarch 17

2011

User Requirements Software Requirements

Umambiguous data formal way of representing information about data

Unambiguous access well-defined interfaces

Transparency of

computational tools

formal way of representing information about

methods well-defined interfaces

Variety of user groups simplicity and modularity of design

Need to integrate various

resources (eg databases

prediction methods

models hellip) to make

meaningful predictions

distributed architecture interoperability

Need to integrate

biological information

again modularity of design extensibility

bull Distributed Web Services for predictive toxicology

bull Several types of Web Services (using REST Web service technology )

bull Service types corresponds to following building blocks

ndash Chemical compound

ndash Dataset of chemical compounds and their properties (calculated or measured)

ndash Algorithm (descriptor calculation regression classification structural alerts

QC etc)

ndash Predictive model

ndash Report validation applicability domain etc

bull Every object (compound dataset algorithm model etc) has an

unique web address ( eg httpmyhostcommodelbestpredictivemodel )

bull These objects can be created read deleted and updated

bull Every object has RDF (W3C Resource Description Framework)

representation defined in OpenTox ontology (opentoxowl)

OpenTox framework

5Ideaconsult LtdMarch 17 2011

Why integration framework for predictive

toxicology

March 17 2011

bull Challengesndash Chemical structures

bull Might be ambiguous

bull Might be error prone or time consuming to reproduce from publications

ndash Data bull Multiple formats

bull Implicit semantics often buried in human readable documentation only

ndash Modelsbull Tens of thousands available in software or in publications

bull Multiple software solutions mostly incompatible

bull Predictions reproducibility is time consuming and often hard to achieve

bull Automatic comparison of prediction results difficult

Why integration framework for predictive

toxicology

March 17 2011 Ideaconsult Ltd3

OpenTox Framework approach

4Ideaconsult LtdMarch 17

2011

User Requirements Software Requirements

Umambiguous data formal way of representing information about data

Unambiguous access well-defined interfaces

Transparency of

computational tools

formal way of representing information about

methods well-defined interfaces

Variety of user groups simplicity and modularity of design

Need to integrate various

resources (eg databases

prediction methods

models hellip) to make

meaningful predictions

distributed architecture interoperability

Need to integrate

biological information

again modularity of design extensibility

bull Distributed Web Services for predictive toxicology

bull Several types of Web Services (using REST Web service technology )

bull Service types corresponds to following building blocks

ndash Chemical compound

ndash Dataset of chemical compounds and their properties (calculated or measured)

ndash Algorithm (descriptor calculation regression classification structural alerts

QC etc)

ndash Predictive model

ndash Report validation applicability domain etc

bull Every object (compound dataset algorithm model etc) has an

unique web address ( eg httpmyhostcommodelbestpredictivemodel )

bull These objects can be created read deleted and updated

bull Every object has RDF (W3C Resource Description Framework)

representation defined in OpenTox ontology (opentoxowl)

OpenTox framework

5Ideaconsult LtdMarch 17 2011

bull Challengesndash Chemical structures

bull Might be ambiguous

bull Might be error prone or time consuming to reproduce from publications

ndash Data bull Multiple formats

bull Implicit semantics often buried in human readable documentation only

ndash Modelsbull Tens of thousands available in software or in publications

bull Multiple software solutions mostly incompatible

bull Predictions reproducibility is time consuming and often hard to achieve

bull Automatic comparison of prediction results difficult

Why integration framework for predictive

toxicology

March 17 2011 Ideaconsult Ltd3

OpenTox Framework approach

4Ideaconsult LtdMarch 17

2011

User Requirements Software Requirements

Umambiguous data formal way of representing information about data

Unambiguous access well-defined interfaces

Transparency of

computational tools

formal way of representing information about

methods well-defined interfaces

Variety of user groups simplicity and modularity of design

Need to integrate various

resources (eg databases

prediction methods

models hellip) to make

meaningful predictions

distributed architecture interoperability

Need to integrate

biological information

again modularity of design extensibility

bull Distributed Web Services for predictive toxicology

bull Several types of Web Services (using REST Web service technology )

bull Service types corresponds to following building blocks

ndash Chemical compound

ndash Dataset of chemical compounds and their properties (calculated or measured)

ndash Algorithm (descriptor calculation regression classification structural alerts

QC etc)

ndash Predictive model

ndash Report validation applicability domain etc

bull Every object (compound dataset algorithm model etc) has an

unique web address ( eg httpmyhostcommodelbestpredictivemodel )

bull These objects can be created read deleted and updated

bull Every object has RDF (W3C Resource Description Framework)

representation defined in OpenTox ontology (opentoxowl)

OpenTox framework

5Ideaconsult LtdMarch 17 2011

OpenTox Framework approach

4Ideaconsult LtdMarch 17

2011

User Requirements Software Requirements

Umambiguous data formal way of representing information about data

Unambiguous access well-defined interfaces

Transparency of

computational tools

formal way of representing information about

methods well-defined interfaces

Variety of user groups simplicity and modularity of design

Need to integrate various

resources (eg databases

prediction methods

models hellip) to make

meaningful predictions

distributed architecture interoperability

Need to integrate

biological information

again modularity of design extensibility

bull Distributed Web Services for predictive toxicology

bull Several types of Web Services (using REST Web service technology )

bull Service types corresponds to following building blocks

ndash Chemical compound

ndash Dataset of chemical compounds and their properties (calculated or measured)

ndash Algorithm (descriptor calculation regression classification structural alerts

QC etc)

ndash Predictive model

ndash Report validation applicability domain etc

bull Every object (compound dataset algorithm model etc) has an

unique web address ( eg httpmyhostcommodelbestpredictivemodel )

bull These objects can be created read deleted and updated

bull Every object has RDF (W3C Resource Description Framework)

representation defined in OpenTox ontology (opentoxowl)

OpenTox framework

5Ideaconsult LtdMarch 17 2011

bull Distributed Web Services for predictive toxicology

bull Several types of Web Services (using REST Web service technology )

bull Service types corresponds to following building blocks

ndash Chemical compound

ndash Dataset of chemical compounds and their properties (calculated or measured)

ndash Algorithm (descriptor calculation regression classification structural alerts

QC etc)

ndash Predictive model

ndash Report validation applicability domain etc

bull Every object (compound dataset algorithm model etc) has an

unique web address ( eg httpmyhostcommodelbestpredictivemodel )

bull These objects can be created read deleted and updated

bull Every object has RDF (W3C Resource Description Framework)

representation defined in OpenTox ontology (opentoxowl)

OpenTox framework

5Ideaconsult LtdMarch 17 2011

OpenTox components

Overview of OpenTox API (Application Programming Interface)

7

Feature

GET

POST

PUT

DELETE

Compound

GET

POST

PUT

DELETE

Dataset

GET

POST

PUT

DELETE

Ontology

GET

POST

PUT

DELETE

Algorithm

GET

POST

PUT

DELETE

Model

GET

POST

PUT

DELETE

AppDomain

GET

POST

PUT

DELETEValidation

GET

POST

PUT

DELETE

Report

GET

POST

PUT

DELETE

bullThe way applications talk to each other

bullThe way developers talk to applications

CompoundData

httpmyhostcomfeature21580

httpmyhostcomfeature21589

httpmyhostcomfeature21573

httpmyhostcomfeature21576

httpmyhostcomfeature21588

httpmyhostcomfeature21858

httpmyhostcomfeature22114

httpmyhostcomcompound413

NN-dimethyl-4-aminoazobenzene

CN(C1=CC=C(C=C1)N=NC2=CC=CC=C2)C

3 331 2253 YES 3123

httpmyhostcomcompound44497

4-

acetamidofl

uorene O=C(Nc3c2c1ccccc1Cc2ccc3)C

1 NP 22328 YES 2085

hellip hellip hellip hellip hellip hellip hellip hellip

Feature

GET

POST

PUT

DELETE

Compound

GET

POST

PUT

DELETE

Dataset

GET

POST

PUT

DELETE

All columns have explicit and machine readable pointers to

originating algorithms models or data

Everything described by W3C RDF (Resource Description framework)

representation

OpenTox datasets Unified access to data

8Ideaconsult LtdMarch 17 2011

httpmyhostcomfeature21573af21573

a otFeature otNumericFeature otNominalFeature

dccreator

httpwwwepagovNCCTdsstoxsdf_isscan_externalhtml

dctitle Canc

othasSource ISSCAN_v3a_1153_19Sept081222179139sdf

= oteeCarcinogenicity

httpmyhostcomfeature21858dctitle Structural Alert for genotoxic carcinogenicity

othasSource

lthttpmyhostcomalgorithmBenigni+2F+Bossa+rul

ebase+28for+mutagenicity+and+carcinogenicitygt

httpmyhostcomfeature22114a otFeature otNumericFeature

dccreator

httpwwwblueobeliskorgontologieschemoinformatics-

algorithmsxlogP

dctitle XLogP

othasSource

lthttpmyhostcomalgorithmorgopensciencecdkqsardescriptors

molecularXLogPDescriptorgt

= oteeOctanol-water_partition_coefficient_Kow

Overview of OpenTox API (Application Programming Interface)

7

Feature

GET

POST

PUT

DELETE

Compound

GET

POST

PUT

DELETE

Dataset

GET

POST

PUT

DELETE

Ontology

GET

POST

PUT

DELETE

Algorithm

GET

POST

PUT

DELETE

Model

GET

POST

PUT

DELETE

AppDomain

GET

POST

PUT

DELETEValidation

GET

POST

PUT

DELETE

Report

GET

POST

PUT

DELETE

bullThe way applications talk to each other

bullThe way developers talk to applications

CompoundData

httpmyhostcomfeature21580

httpmyhostcomfeature21589

httpmyhostcomfeature21573

httpmyhostcomfeature21576

httpmyhostcomfeature21588

httpmyhostcomfeature21858

httpmyhostcomfeature22114

httpmyhostcomcompound413

NN-dimethyl-4-aminoazobenzene

CN(C1=CC=C(C=C1)N=NC2=CC=CC=C2)C

3 331 2253 YES 3123

httpmyhostcomcompound44497

4-

acetamidofl

uorene O=C(Nc3c2c1ccccc1Cc2ccc3)C

1 NP 22328 YES 2085

hellip hellip hellip hellip hellip hellip hellip hellip

Feature

GET

POST

PUT

DELETE

Compound

GET

POST

PUT

DELETE

Dataset

GET

POST

PUT

DELETE

All columns have explicit and machine readable pointers to

originating algorithms models or data

Everything described by W3C RDF (Resource Description framework)

representation

OpenTox datasets Unified access to data

8Ideaconsult LtdMarch 17 2011

httpmyhostcomfeature21573af21573

a otFeature otNumericFeature otNominalFeature

dccreator

httpwwwepagovNCCTdsstoxsdf_isscan_externalhtml

dctitle Canc

othasSource ISSCAN_v3a_1153_19Sept081222179139sdf

= oteeCarcinogenicity

httpmyhostcomfeature21858dctitle Structural Alert for genotoxic carcinogenicity

othasSource

lthttpmyhostcomalgorithmBenigni+2F+Bossa+rul

ebase+28for+mutagenicity+and+carcinogenicitygt

httpmyhostcomfeature22114a otFeature otNumericFeature

dccreator

httpwwwblueobeliskorgontologieschemoinformatics-

algorithmsxlogP

dctitle XLogP

othasSource

lthttpmyhostcomalgorithmorgopensciencecdkqsardescriptors

molecularXLogPDescriptorgt

= oteeOctanol-water_partition_coefficient_Kow

CompoundData

httpmyhostcomfeature21580

httpmyhostcomfeature21589

httpmyhostcomfeature21573

httpmyhostcomfeature21576

httpmyhostcomfeature21588

httpmyhostcomfeature21858

httpmyhostcomfeature22114

httpmyhostcomcompound413

NN-dimethyl-4-aminoazobenzene

CN(C1=CC=C(C=C1)N=NC2=CC=CC=C2)C

3 331 2253 YES 3123

httpmyhostcomcompound44497

4-

acetamidofl

uorene O=C(Nc3c2c1ccccc1Cc2ccc3)C

1 NP 22328 YES 2085

hellip hellip hellip hellip hellip hellip hellip hellip

Feature

GET

POST

PUT

DELETE

Compound

GET

POST

PUT

DELETE

Dataset

GET

POST

PUT

DELETE

All columns have explicit and machine readable pointers to

originating algorithms models or data

Everything described by W3C RDF (Resource Description framework)

representation

OpenTox datasets Unified access to data

8Ideaconsult LtdMarch 17 2011

httpmyhostcomfeature21573af21573

a otFeature otNumericFeature otNominalFeature

dccreator

httpwwwepagovNCCTdsstoxsdf_isscan_externalhtml

dctitle Canc

othasSource ISSCAN_v3a_1153_19Sept081222179139sdf

= oteeCarcinogenicity

httpmyhostcomfeature21858dctitle Structural Alert for genotoxic carcinogenicity

othasSource

lthttpmyhostcomalgorithmBenigni+2F+Bossa+rul

ebase+28for+mutagenicity+and+carcinogenicitygt

httpmyhostcomfeature22114a otFeature otNumericFeature

dccreator

httpwwwblueobeliskorgontologieschemoinformatics-

algorithmsxlogP

dctitle XLogP

othasSource

lthttpmyhostcomalgorithmorgopensciencecdkqsardescriptors

molecularXLogPDescriptorgt

= oteeOctanol-water_partition_coefficient_Kow

bull Ontologies are critical to unambiguously describe data ndash Opentoxowl

ndash Blue Obelisk algorithm ontology (cheminformatics algorithms)

ndash OpenTox algotihm types ontology

ndash OpenTox endpoints ontology based on ECHA endpoints classification

ndash Specific endpoints ontologies developed by OpenTox partners with toxicology

expertise

ndash Existing biomedical ontologies

bull Datasets can be easily merged compared and calculations

reproduced regardless of their physical place

bull OWL sameAs construct can be used to denote two links point to the

same object

bull Authentication and authorization to protect sensitive data

Uniform access to the data

9Ideaconsult LtdMarch 17 2011

bull Datasets can be easily merged compared and calculations

reproduced regardless of their physical place

bull The dataset service offers property compound substructure and

similarity searches via uniform OpenTox Application Programming

Interface

Uniform access to the data

10Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2dataset2344 (the dataset)

bull httpappsideaconsultnet8080ambit2dataset2344metadata

(metadata obviously)

Example mutagenicity dataset

11Ideaconsult LtdMarch 17 2011

bull Datasets can be easily merged compared and calculations

reproduced regardless of their physical place

bull The dataset service offers property compound substructure and

similarity searches via uniform OpenTox Application Programming

Interface

Uniform access to the data

10Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2dataset2344 (the dataset)

bull httpappsideaconsultnet8080ambit2dataset2344metadata

(metadata obviously)

Example mutagenicity dataset

11Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2dataset2344 (the dataset)

bull httpappsideaconsultnet8080ambit2dataset2344metadata

(metadata obviously)

Example mutagenicity dataset

11Ideaconsult LtdMarch 17 2011

Example mutagenicity dataset

12Ideaconsult LtdMarch 17 2011

bull Activity column (httpappsideaconsultnet8080ambit2feature28958)prefix ot lthttpwwwopentoxorgapi11gt

prefix dc lthttppurlorgdcelements11gt

prefix lthttpappsideaconsultnet8080ambit2gt

prefix ota lthttpwwwopentoxorgalgorithmTypesowlgt

prefix otee lthttpwwwopentoxorgechaEndpointsowlgt

hellip

prefix af lthttpappsideaconsultnet8080ambit2featuregt

af28958

a otFeature otNumericFeature

dccreator 1941410136

dctitle Activity

othasSource tox_benchmark_N6512sdf

otunits

= oteeMutagenicity

othasSource

a owlObjectProperty

otunits

a owlDatatypeProperty

otFeature

a owlClass

otNumericFeature

a owlClass

rdfssubClassOf otFeature

Query Is there other mutagenicity data

available

13Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2featuresameas=http3A2F2Fwww

opentoxorg2FechaEndpointsowl23Mutagenicity

httpappsideaconsultnet8080ambit2feature21611

httpappsideaconsultnet8080ambit2feature26221

httpappsideaconsultnet8080ambit2feature28958

httpappsideaconsultnet8080ambit2feature21590

Query Is there other mutagenicity data

available

13Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2featuresameas=http3A2F2Fwww

opentoxorg2FechaEndpointsowl23Mutagenicity

httpappsideaconsultnet8080ambit2feature21611

httpappsideaconsultnet8080ambit2feature26221

httpappsideaconsultnet8080ambit2feature28958

httpappsideaconsultnet8080ambit2feature21590

bull httpappsideaconsultnet8080ambit2dataset2344feature_uris[]=

httpappsideaconsultnet8080ambit2feature28958ampfeature_uris[

]=httpappsideaconsultnet8080ambit2feature21611ampfeature_uri

s[]=httpappsideaconsultnet8080ambit2feature26221ampfeature_u

ris[]= httpappsideaconsultnet8080ambit2feature21590

Merge mutagenicity data

14Ideaconsult LtdMarch 17 2011

Read data from a web address ndash process ndash write to a web address

Uniform access to calculations

15Ideaconsult LtdMarch 17 2011

Feature

GET

POST

PUT

DELETE

Compound

GET

POST

PUT

DELETE

Dataset

GET

POST

PUT

DELETE

Model

GET

POST

PUT

DELETE

+ = Dataset

GET

POST

PUT

DELETE

Predicted results

httpmyhostcomdatasetnewcompounds

httpmyhostcommodelpredictivemodel1

httpmyhostcomdatasetpredictedresults1

List available models

16Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2model

Apply ToxTree module for carcinogenicity

and mutagenicity prediction

17Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2model8

bull The benchmatk dataset web address

bull httpappsideaconsultnet8080ambit2dataset2344

bull Clicking bdquoPredict‟ button initiates calculations

bull It actually sends HTTP POST command to the model URL with parameter

the dataset URL

bull curl ndashX POST ndashd

ldquodataset_uri=httpappsideaconsultnet8080ambit2dataset2344rdquo

httpappsideaconsultnet8080ambit2model8

Read data from a web address ndash process ndash write to a web address

Uniform access to calculations

15Ideaconsult LtdMarch 17 2011

Feature

GET

POST

PUT

DELETE

Compound

GET

POST

PUT

DELETE

Dataset

GET

POST

PUT

DELETE

Model

GET

POST

PUT

DELETE

+ = Dataset

GET

POST

PUT

DELETE

Predicted results

httpmyhostcomdatasetnewcompounds

httpmyhostcommodelpredictivemodel1

httpmyhostcomdatasetpredictedresults1

List available models

16Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2model

Apply ToxTree module for carcinogenicity

and mutagenicity prediction

17Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2model8

bull The benchmatk dataset web address

bull httpappsideaconsultnet8080ambit2dataset2344

bull Clicking bdquoPredict‟ button initiates calculations

bull It actually sends HTTP POST command to the model URL with parameter

the dataset URL

bull curl ndashX POST ndashd

ldquodataset_uri=httpappsideaconsultnet8080ambit2dataset2344rdquo

httpappsideaconsultnet8080ambit2model8

List available models

16Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2model

Apply ToxTree module for carcinogenicity

and mutagenicity prediction

17Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2model8

bull The benchmatk dataset web address

bull httpappsideaconsultnet8080ambit2dataset2344

bull Clicking bdquoPredict‟ button initiates calculations

bull It actually sends HTTP POST command to the model URL with parameter

the dataset URL

bull curl ndashX POST ndashd

ldquodataset_uri=httpappsideaconsultnet8080ambit2dataset2344rdquo

httpappsideaconsultnet8080ambit2model8

Apply ToxTree module for carcinogenicity

and mutagenicity prediction

17Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2model8

bull The benchmatk dataset web address

bull httpappsideaconsultnet8080ambit2dataset2344

bull Clicking bdquoPredict‟ button initiates calculations

bull It actually sends HTTP POST command to the model URL with parameter

the dataset URL

bull curl ndashX POST ndashd

ldquodataset_uri=httpappsideaconsultnet8080ambit2dataset2344rdquo

httpappsideaconsultnet8080ambit2model8

Long running calculations ndash web address

again

18Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2taske75057a0-b005-4c74-

949f-0be9304a3c422model

And the results (a dataset web address)

19Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2dataset2344feature_uris[

]=http3A2F2Fappsideaconsultnet3A80802Fambit22Fmodel2

F82Fpredicted httpappsideaconsultnet8080ambit2

chartpiedataset_uri=httpappsideac

onsultnet8080ambit2dataset2344ampfe

ature_uris[]=httpappsideaconsultnet

8080ambit2feature21858

Building blocks for model creation

20Ideaconsult LtdMarch 17 2011

bull A Datasetndash Datasets can be uploaded (SDF CSV SMI MOL)

bull via HTTP POST to the dataset service

bull Web browser file upload

ndash Existing datasets (uploaded by others) can be used

ndash Dataset Web address assigned

bull Descriptor calculation

bull Learning algorithm

bull Perform predictions

bull Make the model available

The dataset (Caco-2 dataset from a JCIM

publication)

21Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2dataset54

bull httpappsideaconsultnet8080ambit2dataset54metadata

bull The endpoint (caco2 column)

httpappsideaconsultnet8080ambit2feature22200

And the results (a dataset web address)

19Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2dataset2344feature_uris[

]=http3A2F2Fappsideaconsultnet3A80802Fambit22Fmodel2

F82Fpredicted httpappsideaconsultnet8080ambit2

chartpiedataset_uri=httpappsideac

onsultnet8080ambit2dataset2344ampfe

ature_uris[]=httpappsideaconsultnet

8080ambit2feature21858

Building blocks for model creation

20Ideaconsult LtdMarch 17 2011

bull A Datasetndash Datasets can be uploaded (SDF CSV SMI MOL)

bull via HTTP POST to the dataset service

bull Web browser file upload

ndash Existing datasets (uploaded by others) can be used

ndash Dataset Web address assigned

bull Descriptor calculation

bull Learning algorithm

bull Perform predictions

bull Make the model available

The dataset (Caco-2 dataset from a JCIM

publication)

21Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2dataset54

bull httpappsideaconsultnet8080ambit2dataset54metadata

bull The endpoint (caco2 column)

httpappsideaconsultnet8080ambit2feature22200

Building blocks for model creation

20Ideaconsult LtdMarch 17 2011

bull A Datasetndash Datasets can be uploaded (SDF CSV SMI MOL)

bull via HTTP POST to the dataset service

bull Web browser file upload

ndash Existing datasets (uploaded by others) can be used

ndash Dataset Web address assigned

bull Descriptor calculation

bull Learning algorithm

bull Perform predictions

bull Make the model available

The dataset (Caco-2 dataset from a JCIM

publication)

21Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2dataset54

bull httpappsideaconsultnet8080ambit2dataset54metadata

bull The endpoint (caco2 column)

httpappsideaconsultnet8080ambit2feature22200

The dataset (Caco-2 dataset from a JCIM

publication)

21Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2dataset54

bull httpappsideaconsultnet8080ambit2dataset54metadata

bull The endpoint (caco2 column)

httpappsideaconsultnet8080ambit2feature22200

Read data from a web address ndash process ndash write to a web address

Uniform approach to data processing (eg

Descriptors calculation)

22Ideaconsult LtdMarch 17 2011

Feature

GET

POST

PUT

DELETE

Compound

GET

POST

PUT

DELETE

Dataset

GET

POST

PUT

DELETE

Algorithm

GET

POST

PUT

DELETE

+ =

httpmyhostcomdatasettrainingset1

httpmyhostcomalgorithmneuralnetwork

Feature

GET

POST

PUT

DELETE

Compound

GET

POST

PUT

DELETE

Dataset

GET

POST

PUT

DELETE

=

httpmyhostcomdatasetresults

The algorithms (including descriptors

calculation)

23Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2algorithm

Descriptors calculation

24Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2algorithmorgopenscience

cdkqsardescriptorsmolecularXLogPDescriptor

bull The demo dataset web address

bull httpappsideaconsultnet8080ambit2dataset54

bull Clicking bdquoPredict‟ button initiates calculations

bull It actually sends HTTP POST command to the algorithm URL with

parameter the dataset URL

bull curl ndashX POST ndashd

ldquodataset_uri=httpappsideaconsultnet8080ambit2dataset

2344rdquo

httpappsideaconsultnet8080ambit2algorithmorgopensci

encecdkqsardescriptorsmolecularXLogPDescriptor

The algorithms (including descriptors

calculation)

23Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2algorithm

Descriptors calculation

24Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2algorithmorgopenscience

cdkqsardescriptorsmolecularXLogPDescriptor

bull The demo dataset web address

bull httpappsideaconsultnet8080ambit2dataset54

bull Clicking bdquoPredict‟ button initiates calculations

bull It actually sends HTTP POST command to the algorithm URL with

parameter the dataset URL

bull curl ndashX POST ndashd

ldquodataset_uri=httpappsideaconsultnet8080ambit2dataset

2344rdquo

httpappsideaconsultnet8080ambit2algorithmorgopensci

encecdkqsardescriptorsmolecularXLogPDescriptor

Descriptors calculation

24Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2algorithmorgopenscience

cdkqsardescriptorsmolecularXLogPDescriptor

bull The demo dataset web address

bull httpappsideaconsultnet8080ambit2dataset54

bull Clicking bdquoPredict‟ button initiates calculations

bull It actually sends HTTP POST command to the algorithm URL with

parameter the dataset URL

bull curl ndashX POST ndashd

ldquodataset_uri=httpappsideaconsultnet8080ambit2dataset

2344rdquo

httpappsideaconsultnet8080ambit2algorithmorgopensci

encecdkqsardescriptorsmolecularXLogPDescriptor

Results from descriptors calculation (a

dataset web address)

25Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2dataset54feature_uris[]=h

ttp3A2F2Fappsideaconsultnet3A80802Fambit22Fmodel2F92

Fpredicted

The descriptors are stored under the web address

httpappsideaconsultnet8080ambit2feature22114

prefix ot lthttpwwwopentoxorgapi11gt

prefix dc lthttppurlorgdcelements11gt

prefix lthttpappsideaconsultnet8080ambit2gt

prefix ota lthttpwwwopentoxorgalgorithmTypesowlgt

prefix otee lthttpwwwopentoxorgechaEndpointsowlgt

prefix ar lthttpappsideaconsultnet8080ambit2referencegt

prefix bx lthttppurlorgnetnknoufnsbibtexgt

prefix rdfs lthttpwwww3org200001rdf-schemagt

prefix am lthttpappsideaconsultnet8080ambit2modelgt

prefix owl lthttpwwww3org200207owlgt

prefix xsd lthttpwwww3org2001XMLSchemagt

prefix ac lthttpappsideaconsultnet8080ambit2compoundgt

prefix ad lthttpappsideaconsultnet8080ambit2datasetgt

prefix rdf lthttpwwww3org19990222-rdf-syntax-nsgt

prefix ag lthttpappsideaconsultnet8080ambit2algorithmgt

prefix af lthttpappsideaconsultnet8080ambit2featuregt

otAlgorithm

a owlClass

othasSource

a owlObjectProperty

otunits

a owlDatatypeProperty

af22114

a otFeature otNumericFeature

dccreator httpwwwblueobeliskorgontologieschemoinformatics-algorithmsxlogP

dctitle XLogP

othasSource lthttpappsideaconsultnet8080ambit2algorithmorgopensciencecdkqsardescriptorsmolecularXLogPDescriptorgt

otunits

= oteeOctanol-water_partition_coefficient_Kow

otFeature

a owlClass

otNumericFeature

a owlClass

rdfssubClassOf otFeature

lthttpappsideaconsultnet8080ambit2algorithmorgopensciencecdkqsardescriptorsmolecularXLogPDescriptorgt

a otAlgorithm

Merge descriptor with activity ndash dataset

with a web address again

26Ideaconsult LtdMarch 17 2011

httpappsideaconsultnet8080ambit2dataset54feature_uris[]=httpap

psideaconsultnet8080ambit2feature22213ampfeature_uris[]=httpappside

aconsultnet8080ambit2feature22200

httpappsideaconsultnet8080ambit2

chartxydataset_uri=httpappsideaco

nsultnet8080ambit2dataset54ampfeatur

e_uris[]=httpappsideaconsultnet8080

ambit2feature22213ampfeature_uris[]=ht

tpappsideaconsultnet8080ambit2f

eature22200

Read data from a web address ndash process ndash write to a web address

Uniform approach to models creation

27Ideaconsult LtdMarch 17 2011

Feature

GET

POST

PUT

DELETE

Compound

GET

POST

PUT

DELETE

Dataset

GET

POST

PUT

DELETE

Algorithm

GET

POST

PUT

DELETE

Model

GET

POST

PUT

DELETE

+=

httpmyhostcomdatasettrainingset1

httpmyhostcomalgorithmneuralnetwork

httpmyhostcommodelpredictivemodel1

Build a regression model

28Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2algorithmLR

bull The demo dataset web address

bull httpappsideaconsultnet8080ambit2dataset54

bull Clicking bdquoPredict‟ button initiates calculations

bull It actually sends HTTP POST command to the algorithm URL with

parameter the dataset URL

bull curl ndashX POST ndashd

ldquodataset_uri=httpappsideaconsultnet8080ambit2dataset54rdquo ndashd

ldquoprediction_feature=

httpappsideaconsultnet8080ambit2feature22200 ldquo

httpappsideaconsultnet8080ambit2algorithmLR

The model

29Ideaconsult LtdMarch 17 2011

httpappsideaconsultnet8080ambit2model57

Web address of results could be retrieved

httpappsideaconsultnet8080ambit2model57predicted in

this case it is a the web address

httpappsideaconsultnet8080ambit2feature28963

Web address of X variables could be retrieved

viahttpappsideaconsultnet8080ambit2model57independent

in this case it is a the web address of

httpappsideaconsultnet8080ambit2feature22213

Build a regression model

28Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2algorithmLR

bull The demo dataset web address

bull httpappsideaconsultnet8080ambit2dataset54

bull Clicking bdquoPredict‟ button initiates calculations

bull It actually sends HTTP POST command to the algorithm URL with

parameter the dataset URL

bull curl ndashX POST ndashd

ldquodataset_uri=httpappsideaconsultnet8080ambit2dataset54rdquo ndashd

ldquoprediction_feature=

httpappsideaconsultnet8080ambit2feature22200 ldquo

httpappsideaconsultnet8080ambit2algorithmLR

The model

29Ideaconsult LtdMarch 17 2011

httpappsideaconsultnet8080ambit2model57

Web address of results could be retrieved

httpappsideaconsultnet8080ambit2model57predicted in

this case it is a the web address

httpappsideaconsultnet8080ambit2feature28963

Web address of X variables could be retrieved

viahttpappsideaconsultnet8080ambit2model57independent

in this case it is a the web address of

httpappsideaconsultnet8080ambit2feature22213

The model

29Ideaconsult LtdMarch 17 2011

httpappsideaconsultnet8080ambit2model57

Web address of results could be retrieved

httpappsideaconsultnet8080ambit2model57predicted in

this case it is a the web address

httpappsideaconsultnet8080ambit2feature28963

Web address of X variables could be retrieved

viahttpappsideaconsultnet8080ambit2model57independent

in this case it is a the web address of

httpappsideaconsultnet8080ambit2feature22213

Read data from a web address ndash process ndash write to a web address

Uniform access to calculations

30Ideaconsult LtdMarch 17 2011

Feature

GET

POST

PUT

DELETE

Compound

GET

POST

PUT

DELETE

Dataset

GET

POST

PUT

DELETE

Model

GET

POST

PUT

DELETE

+ = Dataset

GET

POST

PUT

DELETE

Predicted results

httpmyhostcomdatasetnewcompounds

httpmyhostcommodelpredictivemodel1

httpmyhostcomdatasetpredictedresults1

Apply the new regression model

31Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2model57 (Model)

bull The demo dataset web address

httpappsideaconsultnet8080ambit2dataset54feature_uris[]=ht

tpappsideaconsultnet8080ambit2feature22213ampfeature_uris[]=

httpappsideaconsultnet8080ambit2feature22200

Apply the new regression model

31Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2model57 (Model)

bull The demo dataset web address

httpappsideaconsultnet8080ambit2dataset54feature_uris[]=ht

tpappsideaconsultnet8080ambit2feature22213ampfeature_uris[]=

httpappsideaconsultnet8080ambit2feature22200

Predictions finally (at a dataset web

address)

32Ideaconsult LtdMarch 17 2011

bull httpappsideaconsultnet8080ambit2dataset54feature_uris[]=h

ttpappsideaconsultnet8080ambit2feature22213ampfeature_uris[

]=httpappsideaconsultnet8080ambit2feature22200ampfeature_u

ris[]=http3A2F2Fappsideaconsultnet3A80802Fambit22Fmodel

2F572Fpredicted

Advanced options

(a bit of command line no web browser)

33Ideaconsult LtdMarch 17 2011

bull httpcurlhaxxse

bull curl is a command line tool for transferring data with URL syntax

bull Copy the dataset to a new onecurl -X POST -d

dataset_uri=httpappsideaconsultnet3A80802Fambit22Fdataset2F543Ffeature_uris5B5D3Dhttp3A2F2Fappsideaconsultnet3A80802Fambit22Ffeature2F2221326feature_uris5B5D3Dhttp3A2F2Fappsideaconsultnet3A80802Fambit22Ffeature2F2220026feature_uris5B5D3Dhttp253A252F252Fappsideaconsultnet253A8080252Fambit2252Fmodel252F57252Fpredictedrdquo httpappsideaconsultnet8080ambit2dataset

The new URL returned is

httpappsideaconsultnet8080ambit2datasetR3940

Advanced options

(a bit of command line no web browser)

33Ideaconsult LtdMarch 17 2011

bull httpcurlhaxxse

bull curl is a command line tool for transferring data with URL syntax

bull Copy the dataset to a new onecurl -X POST -d

dataset_uri=httpappsideaconsultnet3A80802Fambit22Fdataset2F543Ffeature_uris5B5D3Dhttp3A2F2Fappsideaconsultnet3A80802Fambit22Ffeature2F2221326feature_uris5B5D3Dhttp3A2F2Fappsideaconsultnet3A80802Fambit22Ffeature2F2220026feature_uris5B5D3Dhttp253A252F252Fappsideaconsultnet253A8080252Fambit2252Fmodel252F57252Fpredictedrdquo httpappsideaconsultnet8080ambit2dataset

The new URL returned is

httpappsideaconsultnet8080ambit2datasetR3940

Read data from a web address ndash process ndash write to a web address

Uniform approach to models validation and

report generation

34Ideaconsult LtdMarch 17 2011

Dataset

GET

POST

PUT

DELETE

Model

GET

POST

PUT

DELETE

+

=Validation

GET

POST

PUT

DELETE

Report

GET

POST

PUT

DELETEModel generating

predictions

Validation report

httpmyhostcomreport1

httpmyhostcomdatasettrainingset1

httpmyhostcomdatasetpredictedresults1

httpmyhostcommodelpredictivemodel1

httpmyhostcomvalidation

Ask the Validation service at Freiburg

to validate the model

35Ideaconsult LtdMarch 17 2011

bull curl -d

test_dataset_uri=httpappsideaconsultnet8080ambit2datasetR3940

-d

prediction_feature=httpappsideaconsultnet8080ambit2feature222

00 -d model_uri=httpappsideaconsultnet8080ambit2model57

httpopentoxinformatikuni-freiburgdevalidation

Task URL returned httpopentoxinformatikuni-

freiburgdetask4062

Query if the task is completed

bull curl -H Accepttexturi-list httpopentoxinformatikuni-

freiburgdetask4062

bull Yes it is here is the address of the validation object

httpopentoxinformatikuni-freiburgdevalidation18920

Ask the Validation service at Freiburg

to validate the model

35Ideaconsult LtdMarch 17 2011

bull curl -d

test_dataset_uri=httpappsideaconsultnet8080ambit2datasetR3940

-d

prediction_feature=httpappsideaconsultnet8080ambit2feature222

00 -d model_uri=httpappsideaconsultnet8080ambit2model57

httpopentoxinformatikuni-freiburgdevalidation

Task URL returned httpopentoxinformatikuni-

freiburgdetask4062

Query if the task is completed

bull curl -H Accepttexturi-list httpopentoxinformatikuni-

freiburgdetask4062

bull Yes it is here is the address of the validation object

httpopentoxinformatikuni-freiburgdevalidation18920

The validation resource

36Ideaconsult LtdMarch 17 2011

curl httpopentoxinformatikuni-freiburgdevalidation18920

---

regression_statistics

sum_squared_error 664592554978574

sample_correlation_coefficient 042022542118295

root_mean_squared_error 0929035636699631

mean_absolute_error 0752657632127617

r_square 0176589404608387

target_variance_actual 106200222895221

target_variance_predicted 0187523074000816

date 2010-08-03 094148 +0200

test_target_dataset_uri

real_runtime 1

prediction_dataset_uri httpappsideaconsultnet8080ambit2datasetR3940feature_uris[]=http3A2F2Fappsideaconsultnet3A80802Fambit22Fmodel2F572Fpredicted

algorithm_uri httpappsideaconsultnet8080ambit2algorithmLR

percent_without_class 00

num_without_class 0

prediction_feature httpappsideaconsultnet8080ambit2feature22200

percent_unpredicted 00

num_instances 77

training_dataset_uri

model_uri httpappsideaconsultnet8080ambit2model57

num_unpredicted 0

test_dataset_uri httpappsideaconsultnet8080ambit2datasetR3940

The validation report

37Ideaconsult LtdMarch 17 2011

curl -d validation_uris=httpopentoxinformatikuni-freiburgdevalidation18920 httpopentoxinformatikuni-freiburgdevalidationreportvalidation

httpopentoxinformatikuni-freiburgdetask4063 lt---- Task URI

curl httpopentoxinformatikuni-freiburgdetask4063 - Ask if completed

---

description

date 2010-08-03T094831+0200

uri httpopentoxinformatikuni-freiburgdetask4063

hasStatus Completed

title

percentageCompleted 00

resultURI

httpopentoxinformatikuni-freiburgdevalidationreportvalidation160

due_to_time 2010-08-03T104831+0200

creator

The validation report

37Ideaconsult LtdMarch 17 2011

curl -d validation_uris=httpopentoxinformatikuni-freiburgdevalidation18920 httpopentoxinformatikuni-freiburgdevalidationreportvalidation

httpopentoxinformatikuni-freiburgdetask4063 lt---- Task URI

curl httpopentoxinformatikuni-freiburgdetask4063 - Ask if completed

---

description

date 2010-08-03T094831+0200

uri httpopentoxinformatikuni-freiburgdetask4063

hasStatus Completed

title

percentageCompleted 00

resultURI

httpopentoxinformatikuni-freiburgdevalidationreportvalidation160

due_to_time 2010-08-03T104831+0200

creator

The validation report (human readable this time)httpopentoxinformatikuni-freiburgdevalidationreportvalidation160

38Ideaconsult LtdMarch 17 2011

bull Could be hosted on a single computer (even a laptop or netbook would do) o

bull Could be transparently distributed on multiple servers in various physical locations in particular for better reliability resilience and scalability

bull OpenTox webservices could be deployed behind firewalls in Intranets (or even offline) when very tight security policies would have to be met

bull Third parties willing to deploy OpenTox webservices could select a relevant subset of services to run tailored to their specific needs

OpenTox web services implementation

39Ideaconsult LtdMarch 17

2011

Make the model available

40Ideaconsult LtdMarch 17 2011

Register at OpenTox ontology service

ndash RDF tripple storage

ndash Accepts HTTP POST

ndash SPARQL endpoint

Curl ndashX POST ndashd

rdquouri=httpappsideaconsultnet8080ambit2model57

rdquo httpappsideaconsultnet8080ontology

Becomes visible for applications

bull Could be hosted on a single computer (even a laptop or netbook would do) o

bull Could be transparently distributed on multiple servers in various physical locations in particular for better reliability resilience and scalability

bull OpenTox webservices could be deployed behind firewalls in Intranets (or even offline) when very tight security policies would have to be met

bull Third parties willing to deploy OpenTox webservices could select a relevant subset of services to run tailored to their specific needs

OpenTox web services implementation

39Ideaconsult LtdMarch 17

2011

Make the model available

40Ideaconsult LtdMarch 17 2011

Register at OpenTox ontology service

ndash RDF tripple storage

ndash Accepts HTTP POST

ndash SPARQL endpoint

Curl ndashX POST ndashd

rdquouri=httpappsideaconsultnet8080ambit2model57

rdquo httpappsideaconsultnet8080ontology

Becomes visible for applications

Make the model available

40Ideaconsult LtdMarch 17 2011

Register at OpenTox ontology service

ndash RDF tripple storage

ndash Accepts HTTP POST

ndash SPARQL endpoint

Curl ndashX POST ndashd

rdquouri=httpappsideaconsultnet8080ambit2model57

rdquo httpappsideaconsultnet8080ontology

Becomes visible for applications

OpenTox Is A Framework

Framework

Unified Access

Open Source

bull Toxicity data

bull QSAR models

bull Validation support

bull Interpretation aids

bull Toxicologists

bull QSAR Modelers

bull API for new QSAR algorithmdevelopment amp integration

bull To optimise impact

bull To allow inspection review

bull To attract external contributors

OpenTox services can be used to develop specific applications or embedded in

workflow systems

bull Two end user oriented demo applications making use of OpenTox

webservices have been developed deployed and are available for

testing ndash httptoxcreateorg and httptoxpredictorg

bull ToxCreate creates models from user supplied datasets

bull ToxPredict uses existing OpenTox models to estimate

chemicalcompound properties

Demo applications

42Ideaconsult LtdMarch 17

2011

bull Two end user oriented demo applications making use of OpenTox

webservices have been developed deployed and are available for

testing ndash httptoxcreateorg and httptoxpredictorg

bull ToxCreate creates models from user supplied datasets

bull ToxPredict uses existing OpenTox models to estimate

chemicalcompound properties

Demo applications

42Ideaconsult LtdMarch 17

2011

bull Interested in applying predictive models but not in

OpenTox implementation details

bull Interested in building predictive models but not in

OpenTox implementation details

bull End users interested in learning implementation details of

OpenTox web services

bull

bull Developers

43

Exercises (4 options)

httpwwwideaconsultnetdownloadsecheminfo

Thank you

March 17

2011

44Ideaconsult Ltd

Thank you

March 17

2011

44Ideaconsult Ltd