Rough Set Microbiome Characterisation

TL;DR

This paper introduces the novel application of Rough Set Theory to microbiome data analysis, demonstrating its effectiveness in characterising gut microbiomes and uncovering new insights into the microbiome-gut-brain axis in depression.

Contribution

It is the first to apply Rough Set Theory for microbiome characterization, offering a new approach that handles complex data without strong assumptions and aids in knowledge discovery.

Findings

RST accurately characterizes microbiomes in depressed subjects

Identifies new microbiome alterations related to the gut-brain axis

Provides a potential solution for microbiome data normalization

Abstract

Microbiota profiles measure the structure of microbial communities in a defined environment (known as microbiomes). In the past decade, microbiome research has focused on health applications as a result of which the gut microbiome has been implicated in the development of a broad range of diseases such as obesity, inflammatory bowel disease, and major depressive disorder. A key goal of many microbiome experiments is to characterise or describe the microbial community. High-throughput sequencing is used to generate microbiota profiles, but data gathered via this method are extremely challenging to analyse, as the data violate multiple strong assumptions of standard models. Rough Set Theory (RST) has weak assumptions that are less likely to be violated, and offers a range of attractive tools for extracting knowledge from complex data. In this paper we present the first application of RST for characterising microbiomes. We begin with a demonstrative benchmark microbiota profile and extend the approach to gut microbiomes gathered from depressed subjects to enable knowledge discovery. We find that RST is capable of excellent characterisation of the gut microbiomes in depressed subjects and identifying previously undescribed alterations to the microbiome-gut-brain axis. An important aspect of the application of RST is that it provides a possible solution to an open research question regarding the search for an optimal normalisation approach for microbiome census data, as one does not currently exist.