Campus alert status is yellow: For the latest campus alert status, news and resources, visit umassmed.edu/coronavirus

Search Close Search
Page Menu

A new computational framework for the prediction of microbiome dynamics

The success of using mathematical modeling to predict functional microbiome biotherapeutics crucially depends on our ability in accurately inferring dynamical systems models of this community under external perturbations. In collaboration with the Gerber laboratory at Brigham and Women’s Hospital, in this project I developed the Microbial Dynamical Systems INference Engine (MDSINE), an open source software package to infer and predict microbiome dynamics from time-series metagenomics that performs all analysis steps from reading data files through generation of figures (Bucci et al., 2016). MDSINE significantly extended our prior work (Bucci et al., 2012a; Buffie et al., 2015; Stein et al., 2013) by implementing: (1) a new technique for accurate estimation of microbial growth trajectories and gradients, specifically tailored for microbiome sequencing data that may be noisy, and irregularly sampled in time, (2) Bayesian  methods for estimation of confidence in parameters, including connectivity in microbial interaction networks, and (3) biologically realistic constraints on model parameters, and (4) a set of computational methods for analyzing long and short term dynamics of poly-microbial systems (Bucci et al., 2016). Our software provides users with several alternative inference algorithms and analysis options, enabling both exploratory and focused analyses. Specifically, studies that we published in (Bucci et al., 2016) demonstrated the increased performance and accuracy of the MDSINE Bayesian algorithms when compared to our precedent ones (Stein et al., 2013). The paper has been cite46 times and features my Ph.D. student Matt Simmons as fourth-author. Matt was an integral part in the actual software implementation and development. MDSINE is available for public download and use at https://bitbucket.org/MDSINE/mdsine/overview 

The initial development of the MDSINE pipeline has been supported by an Advances in Biological Informatics Innovation NSF grant (DBI 1458347) which was awarded to me in September 2015 (Bucci PI). A large part of this grant comprehends the validation and the “stress-testing” of the underlying inference algorithms on ground-truth time-series data generated from (a) simulations with in silico models of multispecies microbial communities of known structure as well as (b) high-throughput fluorescence microscopy observations of microchemostats seeded with a three- five-species bacterial community with an interactions network set by us via synthetic engineering. For (a) my PhD student Matt Simmons developed a new hybrid agent-based/cellular automaton/differential equation framework to simulate dynamics of polymicrobial biofilms, such as those inhabiting the mucosal surface. This python-based framework was implemented with the specific goal of being able to set up simulations with any arbitrary number of species, solutes and system geometry. We have recently published the framework and its application in the context of phage-bacteria dynamics with the purpose of studying how physical and chemical properties of a biofilm matrix impact ability of phages to eradicate an established microbial biofilm. This work was performed in collaboration with Carey Nadell (Dartmouth College) and Knut Drescher (Max Planck) and was published in Fall 2017 in ISME J (Simmons et al., 2017) Dr. Nadell and I have recently received an NSF-MCB (Bucci Co-PI) to couple my poly-microbial biofilm modeling tool and his microfluidics-based experimental system for biofilm engineering and investigate the ecological dynamics that would emergence from the interaction between phages and biofilm-dwelling bacteria.  

Testing of the microbial interactions inference engines on data gathered using synthetically engineered microbial systems is currently a topic under exploration by another PhD student of mine, Jacob Palmer. To construct networks of known interactions we have synthesized a number of genetic constructs capable of inducible production of different microbial microcins. While building this library we noticed the ability of one of the used microcins in inhibiting gastro-enteritis causing Salmonella species. We therefore decided to redirect Jacob’s research towards exploring the possibility of developing smart engineered probiotics to kill drug-resistant enteric pathogens (see Project #3). Additionally, for this subproject, in collaboration with soil microbiologist Mark Silby (Biology UMass Dartmouth) and metabolic engineer Chris Brigham (Bioengineering UMass Dartmouth, now at WIT), and with support from two UMass Dartmouth seed funding awards we have built networks of interacting soil microbes (Ly et al., 2015) to use as input for the testing of our MDSINE pipeline.   

Despite the major advances in microbiome and statistical modeling provided by our developed tools there are still several issues that needs to be tackled in the field of longitudinal microbiome analysis which include high measurement noise, irregular and sparse temporal sampling, and complex dependencies between variables. In collaboration with Georg Gerber, we received an R01 grant fromNGMS (Bucci Co-PI) is to introduce new capabilities, improve on, and provide state-of-the-art implementations of tools for analyzing dynamics, or patterns of change in microbiome time-series data. The tools we develop use Bayesian machine learning methods, which are well-recognized for their strong conceptual and practical advantages, particularly in biomedical domains. Our tools will be rigorously tested and validated on synthetic and real human microbiome data, including publicly available datasets and those from collaborators providing 16S rRNA sequencing, metagenomic, and metabolomics data.