neuroweb: modeling cerebrovascular phenotypes gianluca colombo daniele merico disco, università di...

NEUROWEB:Modeling cerebrovascular phenotypes

Gianluca ColomboDaniele Merico

DISCo, Università di Milano-Bicocca

SSW09Signs, Symptoms and Findings:

Towards an Ontology for Clinical Phenotypes.

Milan, 4-5 September 2009

European CommissionSixth Framework Programme

Information Society Technologies project number 518513

NEUROWEB Project

Act IAssociation Studies

and Clinical Phenotypes

Association Studies

1. Identify patients with a common (clinical) phenotype

2. Determine the statistical significance of association for– Genotype (i.e. Single Nucleotide Polymorphisms)– Environmental / life style factors– Etc…

Other patients

Phenotype carriers(e.g. Severe Stenosis)

SNP S1

SNP S2

SNP S3

Genotype

Prosthetic Heart Valve

Smoked Cigarettes

Clinical Features and Risk Factors

Blood Pressure

Clinical Phenotype(Association Studies)

• Of clinical interest, i.e. typically encountered as an intermediate or final stage of diagnosis, or treatment formulation, or prognosis

• Clinically abnormal sensu OGMS(?)– Not in the body plan

– Potentially resulting in pain / malaise / dysfunction / death

Clinical Phenotype(Association Studies)

• An aggregate of ‘bodily features’ (sensu OGMS) – directly observabledirectly observable (e.g. brain lesion)

or derived / inferablederived / inferable (e.g Diabetes Mellitus Type-II)– which is likely to be strongly associated to

• A genotype, i.e. set of genomic positions with nucleotide value(s)

• A (set of) physiological parameter (e.g. blood pressure)

• A (set of) environmental or life style factor (e.g. cholesterol-rich diet)

• Aggregation rationale: – StrengthStrength of association

– ParsimonyParsimony of associated factorsgenotype-phenotype

• Mutation in a single gene (strictly mendelian)• Multigenic (but maybe belonging to a certain functional category or pathway)

Hypothesis formulation and testing cycles

• NB: AssociationNB: Association may (or may not) imply causationcausation

Etio-physiological relatednessof aggregated bodily features

Why association studies

• Association

Prognosis / improved diagnostics

• Causation

Part of the research elucidating the triads

disorder – disease – pathological process(es)– prevention

– treatment

NEUROWEB Project

IT system supporting Association Studies Integrating the databases of four cerebrovascular excellence centers

Semantic misalignments among local databases

• Granularity (what level of detail in the diagnostic assessment)

• Methodology (diagnostic process)

NEUROWEB Reference Ontology

+Query system

+Reasoner

Act IIThe NEUROWEB

Reference Ontology

Bottom-Up CDS Building• First Step: Dbs aggregation Union of

the existing clinical repositories;– Identify what is semantically

equivalent but terminologically different

• Second Step: DBs collection cleaning:– Criterion: minimum granularity

• DBs fields constituted of other DBs field have been removed from the CDS;

• The resulting fields corresponds to the minimal diagnostic assessment;

– CDS elements can be:• Signs• Symptoms• Lab Tests• Physical Examination

• Third Step: Higher level concepts definition:

– Clinical Phenotypes aggregation of CDS elements;

Results

CDS

Clinical Phenotypes

Sign

Lab test

Methods

PhysicalExamination

Symptoms

Degree

DiagnosticValue

Top-Down Clinical Phenotype Building

• Design principles:– Represent the way clinicians classify stroke types and subtypes

(Top Phenotype Layer)– Analytically deconstruct the stroke types to grant methodological

consistency and resolve granularity discrepancies(Low Phenotype Layer)

– Ground the stroke type definition on the clinical data via the CDS (Core Data Set Layer)

– DB fields can be mapped both at the CDS and at the Low Phenotypes level (Local Database to Reference Ontology)

• TOAST Classification System (Top Phenotypes)– The TOAST Classification System is adopted by the

NEUROWEB clinical communities– and used by the international community to guide:

• Diagnosis• Treatment• Prevention• Trials to identify best practices for these

(Evidence Based Medicine)

– The TOAST phenotypes need analytical deconstruction

to grant methodological consistency and resolve granularity discrepancies among databases (Low Phenotypes, CDS Elements)

Capital Clinical Phenotypesin the cerebrovascular domain

Top-down CDS Building


• Clinical Phenotype:– Top Phenotype:

• Ischemic Stroke types classified according to different criteria:

– Etiology/Anatomy (Atherosclerotic, Cardioembolic, Lacunar).

CDS

Ischemic Stroke types

PhysicalExamination

Anatomy

SignLab test Symptoms

DiagnosticValue

Top Phenotype

Low Phenotype

Etiology



• Clinical Phenotype:– Top Phenotype– Low Phenotype

• Etiological Layer– Derived Evidences– Direct Evidences

• Anatomical Layer– Anatomical Components– Topological Entities


TopologicalEntity

AnatomicalComponents

Top Phenotype

Low Phenotype

Direct Evidences

Derived Evidences




• Etiological Layer– Derived Evidences

» Durative Background

» Traumatic Point Event

• Direct Evidences• Anatomical Layer

– Anatomical Components– Topological Entities


TopologicalEntity


Top Phenotype

Low Phenotype

Direct Evidences

Durative Background

Traumatic Point Event

Derived Evidences





» Durative Background» Traumatic Point Event

– Direct Evidences» Durative Diagnostic

Evidence» Point-event Diagnostic

Evidence



Point-EventDiagnostic Evidence

DurativeDiagnostic Evidence

TopologicalEntity


Top Phenotype

Low Phenotype

Durative Background


Derived Evidences

Direct Evidences


CDS


PhysicalExamination



TopologicalEntity



DiagnosticValue

Top Phenotype

Low Phenotype







Evidence


Durative Background


Derived Evidences

Direct Evidences


CDS


PhysicalExamination

Durative Background




TopologicalEntity



DiagnosticValue

Top Phenotype

Low Phenotype







Evidence


Direct-Composite Evidences

Derived Evidences

Atomic Evidences


CDS


PhysicalExamination



TopologicalEntity



DiagnosticValue

Top Phenotype

Low Phenotype



• Etiological Layer– Durative Background

» Traumatic Point Event– Durative Diagnostic

Evidence» Point event Evidence


• Biological Process

BiologicalProcess

BiologicalProcessParticipant

Durative Background


Derived Evidences

Top-Down CDS Building

CDS

Atherosclerotic Ischemic Stroke

PhysicalExamination

Atherosclerosis

Ischemic Traumatic Event

RelevantIschemicLesion

SevereStenosis

Coagulation

PAI-1

Left

InternalCarotidArtery


DiagnosticValue

Top Phenotype

Low Phenotype



• Etiological Layer– Durative Background

» Traumatic Point Event– Durative Diagnostic

Evidence» Point event Evidence


• Biological Process

Reasoner(Jena)

N2LConverter

N2LConverter

DB1

PhenotypeConverter

User2

NWInterface

User1

Mediator Mediator

Queries rewriter

OntologyT-Box

QueryEngine

DB2

QueryEngine

Computational Issues

• Generating queries and accessing local repositories requires two tasks: – The elements of the local

repository need to be mapped into the ones in the Reference Ontology and the CDS (N2L);

– The NEUROWEB phenotypes need to be transformed in queries in terms of the reference ontology elements that map to the local repository (Phenotype Converter).

• The net effect is the translation of high-level concepts into regular SQL queries.

Reasoner(Jena)

N2LConverter

N2LConverter

DB1

PhenotypeConverter

User2

NWInterface

User1

Mediator Mediator

Queries rewriter

OntologyT-Box

QueryEngine

DB2

QueryEngine

• Querying the Ontology (Top Phenotypes) the NW Ontology engine returns portions of the Ontology graph of interest for clinicians w.r.t.:– DBs Analysis (NW

Ontology Engine)

SPQRLqueries

NWOntology engine

Mediator

NW Ontology Engine Services

NWGenomic engine

• Querying the Ontology the NW Ontology engine returns portions of the Ontology graph of interest for clinicians w.r.t.:– DBs Analysis:

• Evaluation of the methodological coherence of the federated data bases according to their compliancy to the NW Ontology (exploiting the Low Phenotype layer, i.e. the Composite Evidences Sub-Layer);


CDS

Top Phenotype

Composite Evidences

Low Phenotype

Atomic Evidences

DerivedEvidences

DBsRecords

Reasoner(Jena)

N2LConverter

N2LConverter

DB1

PhenotypeConverter

User2

NWInterface

User1

Mediator Mediator

Queries rewriter

OntologyT-Box

QueryEngine

DB2

QueryEngine

• Querying the Ontology the NW Ontology engine returns portions of the Ontology graph of interest for clinicians w.r.t.:– Genomic Analysis (NW

Genomic Engine)

SPQRLqueries

NWOntology engine

Mediator


NWGenomic engine

• Querying the Ontology (built-in queries) the NW Ontology engine returns portions of the Ontology graph of interest for clinicians w.r.t.:

– Genomic Analysis:• Evaluating two different

phenotypes associated with a common SNP, which may have a direct causative role in phenotype determination;


CDS

Top Phenotype

Composite Evidences

Low Phenotype

Atomic Evidences

DerivedEvidences

SNP

DBsRecords

BioProcesses

Act IIINEUROWEB meets OGMS

TOPPHENOTYPES

TOPPHENOTYPES

LOWPHENOTYPES

Durative Etiological Background


Point-event Diagnostic Evidence

TOPO-ANATOMICAL ENTITIES

Topological Concepts

Anatomical Parts

BIOMOLECULAR ENTITIES

Biological Process

Biological Process

Participant

Diagnostic Values

CDS Indicators

Has-Diagnostic-Evidence

Has-Cause-PointEvent

Has-Cause-

Durative

By-Means-Of

Has-Value

Has-Side

Has-Location

Involves

Has-Participant

Durative Diagnostic Evidence

TOP PHENOTYPES(Ischemic Stroke types)


LOWPHENOTYPES




Has-Cause-

Durative

Etiological /Pathological

ProcessDisorder Disposition

PathologicalProcesses

Diagnosis

Disposition

Atherosclerosis Atherogenesis AtheroscleroticPlaque(s)

Ath. Ischemic Stroke Risk

Ischemic Stroke (Point Event)

DisorderPathological

Process



LOWPHENOTYPES




Has-Cause-

Durative




Pathological Process Disorder Disposition

Diagnosis

Disposition




Plaque-based Thrombogenesis

TravelingThrombus

Cerebral Artery Occlusion

Ischemia


Process



LOWPHENOTYPES




Has-Cause-

Durative




Pathological Process Disorder Disposition

Diagnosis

Disposition





TravelingThrombus


Occluded Vessel

TOPPHENOTYPES

TOPPHENOTYPES

LOWPHENOTYPES



Diagnostic Values

CDS Indicators



Has-Cause-

Durative

By-Means-Of

Has-Value



ProcessDisorderDisposition


Finding

Stenosis


TOPPHENOTYPES

TOPPHENOTYPES

LOWPHENOTYPES



Diagnostic Values

CDS Indicators



Has-Cause-

Durative

By-Means-Of

Has-Value


Pathological Process

Ischemia


Brain tissuedeath


Disorder

Ischemicscar

Finding

Scan lesion


ProcessPathological

Process Disorder

TravelingThrombus


Ischemia


Oxygen-deprivedBrain tissue


Brain tissuedeath

Biomolecular Process

Coagulation

TOPPHENOTYPES

TOPPHENOTYPES

LOWPHENOTYPES




Biological Process

Biological Process

Participant


Has-Cause-

Durative

Involves

Has-Participant


Hypoxia-inducedApoptosis


ProcessPathological

Process Disorder

TravelingThrombus

Ischemia


Oxygen-deprivedBrain tissue


Brain tissuedeath


Coagulation

TOPPHENOTYPES

TOPPHENOTYPES

LOWPHENOTYPES




Biological Process

Biological Process

Participant


Has-Cause-

Durative

Involves

Has-Participant


Hypoxia-inducedApoptosis

How to handle in OGMS the transition from physiology-level to biomolecular-

level processes?

neuroweb: modeling cerebrovascular phenotypes gianluca colombo daniele merico disco, università di...

Documents

neuroweb project slide

common clinical phenotype

cigarettes clinical

existing clinical repositories

phenotype mutation

cds building

patients phenotype carriers

dbs aggregation union