A finite mixture model for X‐chromosome association with an emphasis on microbiome data analysis. Issue 4 (18th January 2019)

• Record Type: Journal Article
• Authors: Espin‐Garcia, Osvaldo; Croitoru, Kenneth; Xu, Wei
• Is Part Of: Genetic epidemiology. Volume 43:Issue 4(2019)
• Page Start: 427
• Page End: 439
• Publication Date: 2019-01-18
• Subjects: finite mixture models -- microbiome data -- random/escape/skewed X‐chromosome inactivation -- X‐chromosome association; Genetic epidemiology -- Periodicals; Heredity -- Periodicals; Medical geography -- Periodicals; 614
• DOI: 10.1002/gepi.22190
• Languages: English
• ISSNs: 0741-0395

Abstract

Abstract: Analysis of the X chromosome has been largely neglected in genetic studies mainly because of complex underlying biological mechanisms. On the other hand, the study of human microbiome data (typically over‐dispersed counts with an excess of zeros) has generated great interest recently because of advancements in next‐generation sequencing technologies. We propose a novel approach to infer the association between host genetic variants in the X‐chromosome and microbiome data. The method accounts for random X‐chromosome inactivation (XCI), skewed (or nonrandom) XCI (XCI‐S), and escape of XCI (XCI‐E). The inference is performed through a finite mixture model (FMM), in which an indicator variable denoting the "true" biological mechanism is treated as missing data. An expectation‐maximization algorithm on zero‐inflated and two‐part models is implemented to estimate genetic effects. We investigate the performance of the FMM along with strategies that assume XCI and XCI‐E mechanisms for all subjects compared with alternative approaches. Briefly, an XCI mechanism codes males' genotypes as homozygous females, whereas under XCI‐E, males are treated as heterozygous females. By comprehensive simulations, we evaluate tests of the hypothesis under a computationally efficient score statistic. In summary, the FMM renders reduced bias and commensurate power compared to XCI, XCI‐E, and alternative strategies while maintaining adequate Type 1 error control. The proposed method has far‐reaching applications. In particular, we illustrate its usage on a large‐scale human microbiome study, the Genetic, Environmental and Microbial (GEM) project, to test for the genetic association on the X chromosome.