Please activate JavaScript!
Please install Adobe Flash Player, click here for download

ISCB2014_abstract_book

26 ISCB 2014 Vienna, Austria • Abstracts - Oral PresentationsMonday, 25th August 2014 • 14:00-15:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust New emerging approaches are the so called entropy-based statistics, sup- ported on the Shannon definition of entropy and the whole information theory. The notions of (conditional) entropy, synergy, (conditional) mutual infor- mation are well-posed and enable to give a measure for gene-gene inter- actions and describe also non-linear dependencies between genotypes and phenotypes. In this presentation we give an overview of the different entropy-based statistics, underlining their strengths and weaknesses. Furthermore, we show that there are more open issues and suggest which directions can be interesting for future work.   C10.2 Network Approach to Identify Gene-by- Secondary Phenotype Interactions in GWAS AN Vidyashankar1 , G Diao1 , B Etain2 , S Katsahian1 1 George Mason University, Fairfax, United States, 2 Institut Universitaire d’Hématologie, Université Paris VII, Paris, France Aims: Analysis of secondary phenotypes in Genome-Wide Association studies has recently received much attention in the literature. Motivated by an application to a recent psychiatric study, the primary objectives of the study are the following: (a) to identify significant associations between the primary phenotype and secondary phenotype by gene interactions and (b) establish an interaction network and use it to identify groups of genes associated with the primary phenotype. Methods: Regression models will be used to estimate gene by secondary phenotype interactions at each marker. An efficient score approach with adjusted p-values is then used to test for the significance of the interaction effects after controlling for family-wise error rate (FWER). These results are then analyzed using network based methods. Network wide metrics are then applied to identify groups of secondary phenotype by gene interac- tions that are associated with the primary phenotype. Results: Use of network wide metrics allows for a principled approach to cluster genes that are significantly associated with the primary phenotype through their interactions with the secondary phenotypes. This facilitates identifying interaction peaks amongst secondary phenotypes. Conclusions: Identification of gene-by-secondary phenotype interac- tions associated with the primary phenotypes helps in identifying differ- ent treatment options for varying subgroups of patients. C10.3 Gene-gene interaction analysis of correlated phenotypes T Park1 , I-S Huh1 , M-S Kwon1 1 Seoul National University, Seoul, Republic of Korea Despite of many successful results from genome-wide association stud- ies (GWAS), only a small number of genetic variants tend to be identified and replicated given a stringent genome-wide significance criterion. Furthermore, in many GWAS, one binary trait is commonly used which is derived from multiple quantitative traits. The use of this summary phe- notype may decrease power due to the loss of information about phe- notypes. Therefore, we propose a multivariate approach which uses all information about correlated phenotypes. Especially, we focus on iden- tification of gene-gene interaction effects on the correlated phenotypes. Generalized multifactor dimensional reduction (GMDR) method has been commonly used in identifying gene-gene interactions. We propose a mul- tivariate GMDR approach in order to identify gene-gene interaction for the multiple phenotypes. Our proposed multivariate GMDR method uses multiple phenotypes simultaneously. We applied the multivariate GMDR approach to a GWA data dataset of 8,842 Korean individuals.   C10.4 Estimating the rediscovery rate for assessing the validation of genome-wide association studies D Lee1 , W Lee2 , A Ganna3 , Y Pawitan3 1 Department of Statistics, Ewha Womans University, Seoul, Republic of Korea, 2 Department of Statistics, Inha University, Incheon, Republic of Korea, 3 Department of MEB, Karolinska Institutet, Stockholm, Sweden Validation studies are often used to confirm that the observed findings in the training studies are not artifact due to chance or uncontrolled bias. These replicated studies increase the generalizability of the results and de- crease the possibility to report false positive findings. Although the impor- tance of replication or validation studies is well recognized, there seems to be less awareness of the factors that influence the reproducibility of signif- icant findings. Indeed, the selection of the validation study is more often driven by data availability rather than study design. In this study we aim to investigate the factors influencing the proportion of significant find- ings from a training sample that are replicated in a validation sample. We quantify this by introducing a measure called rediscovery rate (RDR), and show how to estimate it nonparametrically from the training dataset. This RDR estimate can be used to design and to assess the validation study. Furthermore, we discuss the meaning of local RDR to interpret and mea- sure the reproducibility of each significant SNP outcome. We use simulat- ed data and real examples from genome-wide association studies to illus- trate the application of the RDR and local RDR concept in high-throughput data analyses. C10.5 A multivariate method for meta-analysis of multiple outcomes in genetic association studies NL Dimou1 , PG Bagos1 1 University of Thessaly, Lamia, Greece In this work we present a simple, yet powerful approach for performing multivariate meta-analysis of genetic association studies when multiple outcomes are assessed. The key element of our approach is the analytical calculation of the within-studies covariances. We propose a model based on summary data, uniformly defined for both discrete and continuous outcomes (using log odds-ratios or mean differ- ences). The within-studies covariances can be calculated using the cross- classification of the genotypes in both outcomes, which are retrieved us- ing a log-linear model using the iterative proportional fitting algorithm under the assumption of no three-way interaction. As an example, we assess the association of MDR1 C3435T polymorphism with two exclusive outcomes (Ulcerative colitis and Crohn’s disease), as well as the association of GNB3 C825T polymorphism with two non-exclu- sive dichotomous outcomes (diabetes and hypertension). We also present an application using continuous outcomes (diastolic and systolic blood pressure).We show the applicability and the generality of the method per- forming the analysis assuming the genetic model beforehand or following a genetic model-free approach. The method is simple and fast, it can be extended for several outcomes and can be fitted in nearly all statistical packages. There is no need for individual patient data or the simultaneous evaluation of both outcomes in all studies. We conclude that the proposed method constitute a useful framework for performing meta-analysis for multiple outcomes within the context of ge- netic association studies. Connections to other similar models presented in the literature, are discussed, as well as potential extensions to future work.

Pages Overview