Demetrius DiMucci Successfully Defends His Thesis

Congratulations to Demetrius DiMucci in the successful defense of his thesis, "Machine Learning Applications for Synthetic Microbial Ecology: Predicting Interactions and Identifying Their Putative Mechanisms." Great job Demetrius!

To learn more about Demetrius' thesis, read his abstract below:

ABSTRACT

Microbial communities are key components of Earth’s ecosystems and they play important roles in human health and industrial processes. These communities and their functions can strongly depend on the diverse interactions between constituent species, posing the question of how such interactions can be predicted, measured and controlled. This challenge is particularly relevant for the many practical applications enabled by the rising field of synthetic microbial ecology, which includes the design of microbiome therapies for human diseases. Advances in sequencing technologies and genomic databases provide valuable datasets and tools for studying inter-microbial interactions, but the capacity to characterize the strength and mechanisms of interactions between species in large consortia is still an unsolved challenge. In this thesis, I show how machine learning methods can be used to help address these questions. 

The first portion of my thesis work was focused on predicting the outcome of pairwise interactions between microbial species. By integrating genomic information and observed experimental data, I used machine learning algorithms to explore the predictive relationship between single-species traits and inter-species interaction phenotypes. I found that organismal traits (e.g. annotated functions of genomic elements) are sufficient to predict the qualitative outcome of interactions between microbes. I also found that the relative fraction of possible experiments needed to build acceptable models drastically shrinks as the combinatorial space grows. In the second part of my thesis work, I developed an algorithmic method for identifying putative interaction mechanisms by scoring combinations of variables that random forest uses in order to predict interaction outcomes. I applied this method to a study of the human microbiome and identified a previously unreported combination of microbes that are strongly associated with Crohn’s disease. In the last part of my thesis, I utilized a regression approach to first identify and then quantify interactions between microbial species relevant to community function. The work I present in this dissertation provides a general framework for understanding the myriad interactions that occur in synthetic microbial consortia.

Major Professor: Daniel Segrè