DMS/NIGMS 2: Unraveling the Role of the Human Microbiome to Advance Precision Medicine

Precision medicine seeks to optimize health care quality by tailoring treatments to a person’s unique characteristics. The current practice of precision medicine mainly utilizes individuals’ genomics information. The new sequencing technology allows us to quantify the human microbiome – the full array of microorganisms in the human body.

The rich information in the microbiome holds great potential to advance precision medicine. However, the development of robust and efficient statistical methods that adapt to the unique features of the microbiome data has seriously fallen behind. For instance, the abundance of microbes is measured in fractions and their actual abundance cannot be recovered in sequencing experiments.

Applying standard methods to analyze one microbe at a time will lead to many false discoveries and irreproducible results. In this project, the PIs will develop methods to jointly analyze many microbes for precision medicine. Specifically, the PIs will develop novel statistical methods and computer software to characterize disease pathology and subtypes using microbiome data, nominate microbial biomarkers for personalized diet or drug intake, and identify microbes in the causal pathways of disease progression.

The PIs also plan to provide training to students at all levels, recruit research assistants from under-represented groups, and develop new interdisciplinary courses.

This project aims to develop new statistical methods that are suitable to analyze compositional, high-dimensional, and overdispersed microbiome data for precision medicine applications. In particular, a novel model will be developed to cluster subject-wise microbiome time-series conditional on covariates using the mixture of generalized Dirichlet multinomial distribution.

This mixture model enables researchers to flexibly capture the complex temporal variability of microbiome compositions and produce meaningful disease subtypes. The subject clustering and covariate selection are performed simultaneously, which improves the performance of both analysis tasks. The research will also develop a new framework for learning microbiome-informed personalized treatment rule (PTR) based on the gradient boosting tree method.

Furthermore, a novel knockoff generating method will be used to select the microbe relevant to the PTR with proper false discovery rate control. The developed framework will enable robustly and powerfully selecting microbes that are important for personalized treatment or personalized diet. Lastly, the research will introduce a novel phylogenetic-tree-assisted local mediation model to identify vital mediating taxa for disease progression.

All methods developed in this proposal will be implemented into efficient and user-friendly R packages to advance precision medicine and microbiome research.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.