Generation of a robust reference gut microbiome dataset for an urban population in Argentina optimized by a machine learning approach

Robust human microbiome analysis requires robust reference datasets obtained from a population that presents similar habits to the one we are trying to assess. We reported here the construction of a robust reference dataset of healthy individuals from urban and surrounding rural areas of the Argentine population. We screened 200 volunteers with strict inclusion/exclusion criteria. Volunteers were also screened with routine blood clinical test analysis and a complete metabolome profile from blood and urine to remove outliers before inclusion in the Next Generation Sequencing dataset. Sequencing was done on an Illumina MiSeq using the V3-V4 16S rRNA. Using these data, we performed de novo community structure prediction by applying clustering methodology based on seven distance and dissimilarity metrics and two clustering methods to the reference set. Using this approach, we discovered four different enterotypes in this community structure. We then trained a model for the classification of any new sample into the structure of the reference set. Once the new sample was classified, it was compared to the reference ranges of both the enterotype-specific subset and the whole reference set. Finally, we challenged the robustness of this methodology using samples from two test case volunteers with clinically proven gut dysbiosis in a time-series sampling with dietary interventions. Our results pointed to the need to carefully analyze the results of gut microbiome in the context of enterotype-specific rather than to a whole population dataset.