IIBR Informatics: Mixture model algorithms for inferring covariance structures and microbial associations from microbiome data

Microbial communities are found almost everywhere on earth and they play important functional roles in the environments that they are found in. Microbes in a community interact with each other as they compete for the food and energy resources available in their environment. These direct and indirect interactions between microbes, termed microbial associations, play a large role in determining the structure, organization, and function of the community.

This project addresses the computational challenge of inferring microbial associations from microbiome data generated using high-throughput DNA sequencing technologies. The novel computational tools and resources developed by this project will enable the advancement of knowledge in several disciplines, including environmental sciences, medicine, and human health science.

This project will contribute to understanding the rules of life for microbial ecosystems, and it will further our understanding of the important roles that microbes play in biogeochemical processes in the environment and in the progression of microbe-associated diseases. This project will provide interdisciplinary training for graduate students, with an emphasis on training under-represented groups (including women and minorities).

This project will also contribute to enabling an increased level of high school student participation in STEM areas through the development of an education module that will introduce high-school teachers, via workshops, to introductory topics in genomics and bioinformatics.

Microbial associations can be inferred from the underlying covariance structure that is determined from microbial taxa abundances. These abundances are often estimated from DNA sequence data. However, sequence data are compositional in nature, in the sense that they only provide relative abundance information for taxa, and this poses challenges when determining microbial associations.

Furthermore, associations between groups of microbial taxa are not always fixed, and they can change when factors such as resource availability and environmental characteristics vary. This project will develop novel computational methods to determine the number of covariance structures in large microbiome datasets and to reconstruct the sets of microbial associations.

These methods will be able to capture both positive and negative microbial associations from sequence data while dealing with the challenges posed by the compositional nature of sequence data. The overall approach is based on a mixture model framework incorporating component distributions that model microbial abundance data. This project will develop variational approximation algorithms to determine the number of covariance structures in a given microbiome dataset, fast numerical optimization algorithms to estimate the parameters of the mixture model, and an integrated framework to incorporate metadata in the analysis.

The algorithms will also enable the reconstruction of sparse models, thus handling the scenario when the number of microbial associations in the community is small. The application of these algorithms to analyze large microbiome datasets will generate new insights into microbial ecology of three different environments (human, ocean, and soil). This analysis will include an elucidation of microbial associations at the strain level, the structures of the underlying microbial networks, and the identities of the key taxa in these environments.

The results of the project can be found at https://github.com/syooseph/YoosephLab/tree/master/MixtureMicrobialNetworks.

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.