Kyle Carter's Abstracts

Kyle Carter's Abstracts

Kyle Carter
Ph.D. Student
Statistics

Conferecne Summary
Eastern North American Region
Washington D.C.

 

Lay Abstract

Integrative Analysis for Incorporating Human Microbiome to Improve
Precision Treatment

While it has been known that human DNA and genes can influence a person’s
likelihood of having a disease, recently scientists have discovered that microbes
that live inside the human body can also influence diseases. However, human
DNA can also influence what types of microbes will live inside the body. A special
type of statistical model known as a mediation model can help explain important
information for doctors such as disease state, progression, treatment response,
and more by combining information known about human DNA and how it can
influence microbes, which in turn affect the disease. Current models fail to
include that microbes in the body can correlate to one another. In this research
we propose a new method of selecting important microbes by using the concept
of information entropy, a method that can allow for correlated microbes for small
sample studies. Important microbes and genes for predicting the disease state
are selected using an optimization algorithm. Through a series of comprehensive
simulation studies, the proposed method shows superior performance to current
methods.
 

Abstract

Integrative Analysis for Incorporating Human Microbiome to Improve
Precision Treatment

In recent years, the human microbiome (e.g. in gut), along with the host genome,
has been discovered to play a critical role in disease progression and treatment
effect. However, recent studies have also revealed that host genetic expression
has a marked effect on the composition of species and associated functions in
human microbiota. An integrative –omics study can be performed through
mediation analysis, investigating how the human microbiome mediates the host
gene expression on disease state, progression, treatment response, and more.
Current high dimensional mediation models fail to incorporate the correlation
between mediators (microbial species) or exposures (host genes) without data
transformation which loses biological interpretation. In this research we propose
a novel feature selection approach using a model-based entropy criterion in a
multivariate zero-inflated negative binomial model to allow for correlated
microbial mediators for small sample studies. Mediator microbes genes are
selected in an iterative optimization algorithm. Through a series of
comprehensive simulation studies, the proposed method shows superior
performance to current methods.