Discovering Disease Associations using a Biomedical Semantic Web: Integration and Ranking

One of the principal goals of biomedical research is to elucidate the complex network of gene interactions underlying common human diseases. Although integrative genomics based approaches have been shown to be successful in understanding the underlying pathways and biological processes in normal and disease states, most of the current biomedical knowledge is spread across different databases in different formats. Semantic Web principals, standards and technologies provide an ideal platform to integrate such heterogeneous information and bring forth implicit relations hitherto embedded in these large integrated biomedical and genomic datasets. Semantic Web query languages such as SPARQL can be effectively used to mine the biological entities underlying complex diseases through richer and complex queries on this integrated data. However, the end results are frequently large and unmanageable. Thus, there is a great need to develop techniques to rank resources on the Semantic Web which can later be used to retrieve and rank the results and prevent the information overload. Such ranking can be used to prioritize the discovered disease–gene, disease–pathway or disease–processes novel relationships. We implemented an existing semantic web based knowledge mining technique which not only discovers underlying genes, processes and pathways of diseases but also determines the importance of the resources to rank the results of a search while determining the semantic associations.

Data Integration- RDF MODELData Integration- RDF MODEL

Ranga Chandra Gudivada1,2, Xiaoyan A. Qu 1,2, Anil G Jegga2,3,4, Eric K. Neumann5 , Bruce J Aronow1,2,3,4

Departments of Biomedical Engineering1 and Pediatrics2, University of Cincinnati, Center for Computational Medicine3 and Division of Biomedical Informatics4,Cincinnati Children’s Hospital Medical Center, Cincinnati OH-45229, USA and Teranode Corporation5, Seattle, WA 98104

Case Study-Prioritizing Modifier Genes, Pathways and Biological Processes for Case Study-Prioritizing Modifier Genes, Pathways and Biological Processes for CARDIOMYOPATHY, DILATEDCase Study-Prioritizing Modifier Genes, Pathways and Biological Processes for Case Study-Prioritizing Modifier Genes, Pathways and Biological Processes for CARDIOMYOPATHY, DILATEDAbstractAbstractAbstractAbstract

Computational ProblemComputational ProblemComputational ProblemComputational Problem

Data integration: biological feature complexity is deep, heterogeneous, and extensive.

Data complexity poses a formidable challenge to efforts to integrate, formally model, and simulate biological systems behaviors

Likelihood Ranking requires mining and prioritization of entities and events that function in the context of biological networks

Biological ProblemBiological ProblemBiological ProblemBiological Problem

Disease genes discovered to date likely represent the easy ones. Discovering the genetic basis of remaining Mendelian and complex gene-X-gene-X-environment disorders will be challenging and require consideration of many more features and causal relationships

No gene operates in vacuum, all gene, protein, pathway interactions can lead to Modifier Gene effects

Identifying modifier genes, i.e. gene networks underlying diseases is challenging (pathways, biological processes and functions)

Benefits of Semantic WebBenefits of Semantic WebBenefits of Semantic WebBenefits of Semantic Web

Semantic Web standards such as Resource Description Framework (RDF) & Ontology Web Language (OWL) facilitate semantic integration of heterogeneous multi-source data

SPARQL, a semantic web query language , capable of making queries of higher order relationships in multi dimensional data can be used to mine Bio-RDF graphs

Prioritization of biological entities on semantic web can be accomplished by extending[2] and applying existing graph algorithms, such as Kleinberg Aglorithm[1]

Cell.ComponentGO ID

DiseaseCUI

GeneSymbol

Mol.FunctionGO ID

PathwayId

Biol.ProcessGO ID

Biol.ProcessDescription

Anatomy CUIDisease

Name

Anatomy Name

Mol.FunctionDescription

PathwayDescription

Cell.ComponentDescription

rdfs

:lab

el

rdfs:label

rdfs

:lab

el

rdfs:la

bel

rdfs:label

rdfs:label

inBiological

Process

inMolecula

rFunction

occursIn

Pathway

hasAssociatedGene

ha

sA

ss

oc

iate

dA

na

tom

y

hasAssociated

Disease

Mouse PhenotypeID

Mouse PhenotypeDescription

hasMouse

PhenoType

rdfs

:labe

l

Ranking on Semantic WebRanking on Semantic Web

BIND

REACTOME

Nature Pathway Interaction database

KleinBerg Algorithm (1)

Hig

h A

uth

oritative sco

re

Au

tho

ritative no

de

Pointed by good hubs its authoritative score increasesH

igh

Hu

b s

core

Hu

b N

od

es

Points to many authoritative sites, increases the hub scores

Extending ‘KleinBerg Algorithm’(2) for Semantic Web

gene Pathway

associatedPathway

Objectivity weight

SubjectivityWeight

Subjectivity weight > objectivity weight

A single gene participating in multiple biological pathways is considered more sensitive to perturbation than a single pathway having a large number of nodes (Different weights for non - symmetric properties); corollary :

geneA geneB

interacts

Objectivity weight

SubjectivityWeight

Subjectivity weight = objectivity weight

GeneA interacting with various genes has

equal significance as GeneB interacting with

various genes (Equal weights for symmetric

properties)

CARDIOMYOPATHY,

DILATED,

X-LINKED

Primary Genes

(1)

DMD

Pathways

(1)

Interacting

Partners

(16)

Biological Processes

(4)

Primary genes

+

Interacting Partners

(1+16)

Pathways

(28)

Biological Processes

(27)

Biological Process

GO_0006936 muscle contraction

GO_0007016 cytoskeletal anchoring

GO_0043043 peptide biosynthesis

GO_0007517 muscle development

h_agrPathwayAgrin in Postsynaptic Differentiation

Pathways

QUERY RESULTWITH

PRIORITIZATION

Step1

Step2

Modifier Genes (16)

1 h_agrPathway Agrin in Postsynaptic Differentiation 1.1349842422 h_hsp27Pathway Stress Induction of HSP Regulation 0.1398879183 h_actinYPathway Y branching of actin filaments 0.0939089763 h_no1Pathway Actions of Nitric Oxide in the Heart 0.0939089763 h_nfatPathway NFAT and Hypertrophy of the heart (Transcription in the broken heart)0.0939089763 h_metPathway Signaling of Hepatocyte Growth Factor Receptor0.0939089763 h_salmonellaPathwayHow does salmonella hijack a cell 0.0939089763 h_mCalpainPathway mCalpain and friends in Cell motility 0.0939089763 h_PDZsPathway Synaptic Proteins at the Synaptic Junction 0.0939089763 h_rabPathway Rab GTPases Mark Targets In The Endocytotic Machinery0.093908976

Pathways (28)

1 GO_0006936 muscle contraction 1.53858592 GO_0007517 muscle development 0.35627623 GO_0007165 signal transduction 0.11394034 GO_0048741 skeletal muscle fiber development 0.11029094 GO_0030240 muscle thin filament assembly 0.11029094 GO_0043043 peptide biosynthesis 0.10279024 GO_0007016 cytoskeletal anchoring 0.1027902

Biological Processes (27)

OMIM

Mammalian Phenotype

Others

Disease

Entrez Gene

SwissProt

Gene Ontology

others

Gene / Protein

Annotations

BIOCARTA

KEGG

BIOCYC

Pathways

Molecular

Interactions

PREFIX CCHMC:<http://www.cchmc.com/test.owl#>

PREFIX rdf:<http://www.w3.org/1999/02/22-rdf-syntax-ns#>

SELECT DISTINCT ?pathway

where {

?pathway rdf:type CCHMC:Pathway .

?resource ?PROPERTY ?pathway .

}

SPARQL QUERY

1.Kleinberg, J. M. 1999. Authoritative sources in a hyperlinked environment. J. ACM 46, 5 (Sep. 1999)

2 Bhuvan Bamba, Sougata Mukherjea: Utilizing Resource Importance for Ranking Semantic

Web Query Results. SWDB 2004: 185-198

ConclusionConclusionConclusionConclusion

We have shown that related yet heterogeneous information can be integrated using RDF-OWL and that this approach can support mechanistic analyses of diseases. Specifically, we have uncovered additional genes and pathways that could play a role in the onset and treatment of Cardiomyopathy. We intend to expand our analyses into additional modalities such as anatomy, cellular type, and symptoms/ phenotypes.