MITRE: inferring features from microbiota time-series data linked to host status

Abstract

Longitudinal studies are crucial for discovering causal relationships between the microbiome and human disease. We present MITRE, the Microbiome Interpretable Temporal Rule Engine, a supervised machine learning method for microbiome time-series analysis that infers human-interpretable rules linking changes in abundance of clades of microbes over time windows to binary descriptions of host status, such as the presence/absence of disease. We validate MITRE’s performance on semi-synthetic data and five real datasets. MITRE performs on par or outperforms conventional difficult-to-interpret machine learning approaches, providing a powerful new tool enabling the discovery of biologically interpretable relationships between microbiome and human host ( https://github.com/gerberlab/mitre/ ). Electronic supplementary material The online version of this article (10.1186/s13059-019-1788-y) contains supplementary material, which is available to authorized users.

Scalable learning of interpretable rules for the dynamic microbiome domain

“…As noted, phylogenetic structure is important for microbiome data analysis and various models have been developed that take into account phylogeny (e.g., (Tang et al, 2017)), but these methods are generally not applicable to time-series data. Our work is directly inspired by MITRE (Bogart et al, 2019), which is currently the stateof-the-art for microbiome time-series specific classification methods. However, MITRE's MCMC-based inference algo-rithm suffers from scalability issues, effectively sampling over a discrete combinatoric space of phylogenetic subtrees, time windows and abundance thresholds that must be preenumerated.…”

“…We evaluated the performance of our model on human microbiome data sets originally assessed by MITRE (Bogart et al, 2019), using the same preprocessing and class labels as in the original study, in order to provide comparable results. These data sets consist of 16S rRNA amplicon sequencing data, which identifies the relative abundances of OTUs in the sample.…”

“…We report the mean and standard deviation of F1 score for our model on 10 random seeds, with the runtime given in parenthesis. Scores and runtime of MITRE are taken directly from (Bogart et al, 2019), which provided only mean performance statistics. = Russian vs. Estonian/Finnish nationality) tracked the gut microbiome composition from birth to 3 years of age in cohorts of children at high risk for autoimmune disease.…”

“…A state-of-the-art method in the microbiome field that addresses the above-mentioned challenges is MITRE (Bogart et al, 2019), which uses a fully Bayesian model that learns human-interpretable rules to classify the host's status (e.g.,…”

Predicting microbiomes through a deep latent space

“…We distinguish two major types of study. First, work on single cases, where a unique dataset or problem is addressed, for example, microbial composition being used to predict productivity in soil (Chang et al, 2017), contaminants and geochemical features in wells (Smith et al, 2015), presence/absence of disease due to changes in abundances of microbes over time (Bogart et al, 2019), or biomarkers of cancer (and the type of cancer) from the human blood microbiome (Poore et al, 2020). Second, more general studies are emerging, where multiple datasets of different origin are addressed together, applying the same prediction procedure or tool.…”