Identification of fever and vaccine- associated gene interaction - PowerPoint PPT Presentation

Identification of fever and vaccine- associated gene interaction networks using ontology-based literature mining Arzucan Özgür Bogazici University Junguk Hur, Zuoshuang Xiang, and Yongqun Oliver He University of Michigan The VDOSME workshop, ICBO 2012 July 21, 2012

Motivation

Fever  Fever is a symptom of abnormal elevation of body temperature, usually as a result of a pathologic process.  Fever-associated genes include PGE2, PLA2, COX-2, PTGES, and many cytokines  Many vaccines cause fever, but how vaccination perturbs which fever-related genes is unclear  Goal: Identify gene-gene and gene-vaccine interaction networks that are associated with fever processes using ontology-based literature mining

Workflow

Fever-related literature-derived network

Literature Corpus  Fever-related articles obtained from PubMed:  “Fever OR Hyperthermia OR Pyrexia OR Febrile OR Pyrexial” → 179,156 articles  Vaccine and fever-related articles:  including the terms “vaccine”, “vaccination”, and their variants (e.g., “vaccines”) → 6,224 articles  including 186 specific vaccine names from VO 6,537 articles →  Sentences of titles and abstracts obtained from:  BioNLP database in the National Center for Integrative Biomedical Informatics (NCIBI; http://ncibi.org/)

Vaccine Ontology Support - Motivating Example • These results suggest that the BCG-CWS induces TNF-alpha secretion from DC via TLR2 and TLR4 and that the secreted TNF- alpha induces the maturation of DC per se. [PMID: 11083809] – The term “vaccine” or its variants does not occur in the abstract. – Bacillus Calmette-Guérin (BCG) is a licensed tuberculosis vaccine to protect against infection of Mycobacterium tuberculosis. We use vaccine ontology for two purposes: 1) Obtain vaccine-related literature. 2) Identify specific vaccine-gene interactions.

Vaccine Ontology (VO)  Ontology of the vaccine domain for vaccine data standardization, integration, and analysis.  http://www.violinet.org/vaccineontology/  Classifies a large number of existing vaccines (> 1,000 vaccines) in licensed use, on trial, or in research.  Follows the OBO Foundry principles.  Led by Yongqun “Oliver” He (co-author of this paper).

VO Terms Obtained from http://www.ontobee.org/

Gene and Vaccine Name Identification  Gene names tagged using SciMiner (Hur et al., Bioinformatics, 2009)  Dictionary and rule-based system  F-score: 76%  Genes reported in terms of the official human genes based on the HUGO Gene Nomenclature Committee database (http://www.genenames.org/).  VO-SciMiner used to identify vaccine names based on a set of 186 VO terms (Hur et al., BMC immunology, 2011)  F-score: 95%

Interaction Extraction IL-2 and IL-15 induced the production of IL-17 and IFN-gamma in a dose dependent manner by PBMCs. Path between proteins: good description of No semantic relation between interaction. them. No interaction interaction. Stanford Parser is used to generate the dependency parse trees (de Marneffe et al., 2006).

Path Edit Kernel  Minimum number of operations (insertion, deletion, or substitution of a single word) to transform the first string to the second.  IL2 – nsubj – induced – dobj – production – prep_of – IL-17  IL2 – nsubj – induced – dobj – production – prep_of – IL-17 – conj_and – IFN-gamma  IL-17 – conj_and – IFN-gamma  Edit distance (Path1 -> Path2) = 2 (2 insertions)  Edit distance (Path1 -> Path3) = 8 (6 deletions + 2 insertions) Convert to Similarity Function: EditSim  p i ,p j  =e [ − γ  EditDist  p i ,p j   ] • Integrate as a kernel function to SVM light package (Joachims, 1999). • 56% F-score on AIMED, 85% F-score on CB) (Erkan et al., EMNLP, 2007; Ozgur et al., Journal of Biomedical Semantics, 2011).

Gene-gene interaction networks Gene-gene interactions in all fever- related articles Articles containing the term “vaccine” and its variants Articles containing the term “vaccine” and its variants + terms in the Vaccine Ontology (VO)

Generic fever-related network

Vaccine/VO-associated fever-related network

Centrality Analysis

Degree Centrality  The number of nodes a given node is connected to n k i = ∑ A ij j= 1 z x y  Measures the extent of inluence a node has on the network  The more neighbors a node has, the more important it is  Degree centrality of x = 5; of y = 2

Eigenvector Centrality  Proportional to the sum of the centralities of the neighbors of a given node. n x i = λ − 1 ∑ A ij x j j= 1  In matrix representation: λ x = Ax  For non-negative centrality vector:  λ is largest eigenvalue of A and x is the corresponding eigenvector  Not all neighbors contribute equally to the centrality of node  Defined as “prestige” in social networks  The prestige of a person depends not only on how many friends he has, but also on who (how prestigious) his friends are

Closeness Centrality  Inverse sum of the geodesic distances from a given node to the other nodes in the network closeness  i  = [ ∑ d ij ] − 1 n j= 1 x y  The closer a node to the other nodes, the more important it is . Geodesic distance: length of shortest path between node i and node j (d ij )

Betweenness Centrality  For a node i : sum over all pairs of nodes of proportion of the number of shortest paths passing through i Betweenness  i  = ∑ g jk  i  / g jk g jk (i): # of geodesics passing over i g jk : total # of geodesics j<k x y  Control of a node over the information flow of the network  A node is important if it is on many geodesics

Genes that rank high in both networks - 7 genes - well studied in both contexts

Genes that rank high in generic fever network - 7 genes - not well studied in vaccine context

Genes that rank high in vaccine/VO fever network - 7 genes - well studied in vaccine context

Gene Set Enrichment Analysis

Gene Set Enrichment Analysis  The Database for Annotation, Visualization and Integrated Discovery (DAVID) used.  997 significantly over-represented functional terms (GO or KEGG) in the fever network.  239 significantly over-represented functional terms (GO or KEGG) in the VO-associated fever network.  New scientific hypothesis can be generated (e.g. Role of phosphorylation process in vaccine-induced fever response).

Top 10 most significantly enriched biological functions Values are –log 10 (Benjamini-Hochberg corrected P-values)

Gene-Vaccine Interaction Network

Gene-Vaccine Interactions  SVM pipeline applied to extract gene-vaccine interactions.  1,716 articles containing 2,835 interactions identified. – 32 articles also related to fever – 52 sentences with 44 unique interactions identified. – Specific vaccines: B rucella vaccine RB51, Shigella flexneri vaccine S602, Shigella sonnei strain WRSS1, and Shigella dysenteriae 1 strain WRSd1.  New scientific hypothesis can be generated.

Fever-related gene-vaccine interaction network Green: vaccine Red: genes Blue: genes associated with vaccines New hypothesis: e.g., TLR-vaccine interactions in inducing fever response.

Conclusion  Identification of fever and fever–vaccine associated gene-gene and gene-vaccine networks  Improved mining performance by VO  Phosphorylation-focused regulation enriched in the fever vaccine-subnetwork suggests its crucial role  Identification of TLRs as potential key factors in vaccine-induced fever responses.

Future works  Expansion of the networks by  including more specific vaccines (via improved VO)  using a sentence-level co-citation rather than SVM-based approach  integrating Ontology of Adverse Events (OAE; http://www.oae-ontology.org)  Apply the ontology-based literature mining approach to different domains.

Acknowledgments University of Michigan  Dragomir Radev  Junguk Hur  Eva Feldman  Yongqun He  Alex Ade  Zoushuang Xiang  Brian Athey  Rebecca Racz Funding: NIH grant R01AI081062 & Marie Curie Career Integration Grant within the 7th European Community Framework Programme.

Thank you!