Activity Number:
|
695
|
Type:
|
Contributed
|
Date/Time:
|
Thursday, August 4, 2016 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Section on Statistical Graphics
|
Abstract #321313
|
|
Title:
|
Causal Inference in Observational Studies using data integration in large scales and creating strong instrumental variables
|
Author(s):
|
Azam Yazdani* and Akram Yazdani and Ahmad Samiei and Eric Boerwinkle
|
Companies:
|
and The University of Texas Health Science Center at Houston and Hasso-Plattner-Institut für Softwaresystemtechnik and The University of Texas Health Science Center at Houston
|
Keywords:
|
data integration ;
strong instrumental variable ;
causal inference ;
multi-omics ;
Bayesian network ;
large scale data
|
Abstract:
|
Understanding causal relationships among large numbers of variables is a fundamental goal of a variety of applied domains which can be formalized by Directed Acyclic Graphs (DAGs) and allow one to identify confounders and make optimal inference. The specific characteristic of DAGs is directionality which distinguishes causal relationships from partial correlations. Identification of directionality is not straight-forward due to the Markov equivalence. Application of instrumental variables (IVs) is an approach to identify causal relationships. Weak IVs however increases bias so that estimated directions are not robust. We use genome granularity directed acyclic graph (GDAG) which extract near complete information from the genome granularity to create strong IVs and identify causal relationships among variables of interest in an upper granularity. Our aim was to infer a causal network among 122 metabolites. The dataset consists of 1,034,945 genetic variants scattered across the genome. We applied the GDAG algorithm to identify the metabolomics causal network. Analysis of the network resulted in valuable information discoveries at metabolomics granularity.
|
Authors who are presenting talks have a * after their name.