Thursday, April 15 2021
9:00am - 10:30am
PhD Thesis Presentation
A novel method emphasizing simultaneous associations between correlated phenotypes and a genetic mutation to discover candidate pleiotropy

The phenomenon that one genetic variant simultaneously influences a number of different phenotypes is called pleiotropy. Knowing about pleiotropy gives researchers potential to expand their understanding of the medical impact of a gene. The joint analysis of multiple phenotypes can increase accuracy when predicting a disease. Overall, the more we know about pleiotropy, the more benefit humans can get in terms of prevention, diagnosis, and treatment.
With the main purpose of identifying multiple correlated phenotypes influenced by a variant, we look for genetic associations from the phenotype point of view. This means the phenotypes will switch into the roll of genome in genome-wide association study (GWAS). Instead of using the whole phenome, or all available phenotypes, we can restrict to only a subset of it based on the belief that diseases of the same category are correlated, and correlated phenotypes have a better chance for simultaneous associations with the same variant.
In this dissertation, we first examined the ability of traditional regression methods in identifying pleiotropy from the phenotype point of view, by performing a simulation study on multiple Bernoulli distributed data sets, we compared power and Type 1 error obtained in different simulated scenarios. The two chosen methods are generalized linear models (GLM) and generalized estimating equations (GEE) representing a univariate approach and a multivariate approach, respectively. For the univariate approach, p-values obtained from multiple hypothesis testing were corrected by the Bonferroni correction, simpleM correction, and the sequential test for pleiotropy. These simulations revealed the limits of existing methods for finding candidate pleiotropy.
We then developed the novel weighted penalized network emphasizing associations (WPNEA) method for identifying candidate pleiotropy. WPNEA was built from the mixed graphical models, but the interaction between each pair of two phenotypes is penalized differently based on the interactions of these two phenotypes and the variant. WPNEA pulls out the key traits and gradually adds more phenotypes into the candidate group. Simulation studies showed that in 70 over 72 simulated scenarios, WPNEA obtained more than 90 percent power in identifying clusters with three or four correct traits, and more than 60 power in accepting clusters with all correct traits. When applying to a non-Hispanic White of the COPDGene data, WPNEA replicated a new discovery of a variant on the AGER gene in the association with the emphysema predominant disease, a disease defined as the linear combination of 22 traits. For two more variants on chromosome 15 previously known associated with COPD, WPNEA further suggested a previously unknown association with the airway predominant disease.
Speaker:Minh Chau Nguyen
Location:See Zoom link from email

Download as iCalendar