A Framework of Knowledge Integration and Discovery for Supporting Pharmacogenomics Target Predication of Adverse Drug Events

A Case Study of Drug-Induced Long QT Syndrome

Guoqian Jiang, M.D., Ph.D. Mayo Clinic

2013 AMIA Summit on Translational Bioinformatics March 20, 2013

2013 MFMER | slide-1

Acknowledgements
Co-authors Chen Wang, PhD Qian Zhu, PhD Christopher G. Chute, MD. Dr. PH This work was supported in part by the Pharmacogenomic Research Network (NIH/NIGMS - U19 GM61388 ) and the SHARP Area 4: Secondary Use of EHR Data (90TR000201)

2013 MFMER | slide-2

Introduction
Adverse drug events (ADEs) have been well recognized as a cause of patient morbidity and increased health care costs in the United States. The genetic component of ADEs is being considered as a significant contributing factor for drug response variability and drug toxicity, representing a major component of the movement to pharmacogenomics and individualized medicine
2013 MFMER | slide-3

Identifying pharmacogenomics evidence

Text mining of published literature MEDLINE abstracts Human-based curation approach PharmGKB Binary model Drug-gene Disease-gene Ternary model Drugs vs. ADEs vs. Gene Targets Guiding pharmacogenomics knowledge integration and discovery
2013 MFMER | slide-4

ADEpedia
A standardized knowledge base of ADEs for drug safety surveillance FDA Structured Product Labeling (SPL), FDA Adverse Event Reporting System (AERS) and the Unified Medical Language System (UMLS) http://adepedia.org

2013 MFMER | slide-5

Semantic MEDLINE
It is a recent development by the National Library of Medicine that is a semantically annotated literature corpus. It identifies genes noted in biomedical text as associated with a disease process or a drug and can potentially simplify secondary database curation.

2013 MFMER | slide-6

Objectives
To build a framework of knowledge integration and discovery that aims to support pharmacogenomicstarget predication of ADEs. For knowledge integration we leverage scalable semantic web technologies For knowledge discovery we develop a network analysis-based knowledge discovery approach A case study of long QT syndrome induced by tricyclic antidepressive agents.

2013 MFMER | slide-7

Long QT Syndrome
It is a heart condition in which delayed repolarization of the heart following a heartbeat causes prolongation of the QT interval, and increases the risk of torsades de pointes, ventricular fibrillation and sudden cardiac death. Drug-induced QT prolongation is an increasing public health problem. Many non-cardiac drugs such as tricyclic antidepressants have also been reported to cause QT prolongation.

2013 MFMER | slide-8

System Architecture

2013 MFMER | slide-9

Severe ADE Extraction

Used a normalized dataset extracted from FDAAERS database Putative drug-ADE pairs Outcome codes Validated using ADE dataset from SIDER and the UMLS Classified by the Common Terminology Criteria for Adverse Event (CTCAE) grading system grade >=3 is considered as Severe

2013 MFMER | slide-10

Severity Classification of ADEs

2013 MFMER | slide-11

SPARQL-Enabled Genetic Association Extraction

Long QT Syndrome|C0023976
2013 MFMER | slide-12

SPARQL Query Results

2013 MFMER | slide-13

A Network Analysis-based Approach

Used the Human Protein Reference Database (HPRD) proteinprotein interaction (PPI) network Assumption: closeness of interactions within PPI implies relevance between drug- and ADE- genes calculated average distance of genes between two groups used a random permutation approach for statistical significance prioritized closely related genes Performed Gene Functional Classification and Enrichment analysis an online bioinformatics application known as DAVID developed by the National Institute of Allergy and Infectious Diseases (NIAID), NIH Manually reviewed the relevant PubMed abstracts for validation

2013 MFMER | slide-14

Prioritized drug- and ADE- genes

2013 MFMER | slide-15

Gene Functional Classification

2013 MFMER | slide-16

Validation by a manual review

We retrieved the PubMed IDs linked with the drug genes (KCNK3, ACCN2, KCNJ6 and KCNH2), and manually reviewed all the original abstracts. Among 16 PubMed IDs retrieved, 12 abstracts are true positive (75%), i.e. correctly reflecting the association between a tricyclic antidepressant and a target gene. Of the 12 abstracts, 8 abstracts (linked with KCNH2 ) mentioned of the target drug genes that are related to long QT syndrome whereas 4 abstracts (linked with ACCN2 and KCNJ6) did not mention of. The results indicated that KCNH2 is a well-studied gene across the target drug and the target ADE while ACCN2 and KCNJ6 are potential candidates for the pharmacogenomics-target predication.

2013 MFMER | slide-17

Discussion Knowledge Integration

We integrated a semantically annotated literature corpus Semantic MEDLINE with a normalized ADE knowledgebase ADEpedia using a semantic web-based data integration approach. Bio2RDF (http://bio2rdf.org/ ) Data2Semantics (http://aers.data2semantics.org/) Our ADEpedia project mainly focused on the standardization of ADE knowledge using standard drug and ADE terminologies (e.g., RxNorm, MedDRA, SNOMED CT and UMLS).

2013 MFMER | slide-18

Discussion Knowledge Integration

We found that the normalization of the drugs and the ADEs using the UMLS is extremely important for both data integration and aggregation. For example, the UMLS enabled us to retrieve all the descendants of the drug class Antidepressive Agents, Tricyclic|C0003290, which provided the aggregation power for collecting genetic associations of the drug class.

2013 MFMER | slide-19

Discussion Knowledge Discovery

As genes derived from text mining could be very general and contain many false-positives, we proposed to apply network analysis to filter out less-relevant genes through additional evidence. We will explore other knowledge resources (e.g., Pathway interaction database) to improve the performance of the knowledge discovery model.

2013 MFMER | slide-20

Questions & Discussion

2013 MFMER | slide-21