Application of machine learning techniques for creating urban microbial fingerprints.

Area of Science:

• Microbiology
• Bioinformatics
• Urban Ecology

Background:

• Human-associated microbial communities impact health, but urban biomes are understudied.
• The Metagenomics and Metadesign of the Subways and Urban Biomes (MetaSUB) Consortium studies urban microbiomes.
• The goal is to improve mass transit system design and understand urban microbial fingerprints.

Purpose of the Study:

• To create distinct urban microbial fingerprints for cities.
• To validate the ability to classify unknown urban microbiome samples.
• To identify key microbial species predictive of a sample's city of origin.

Main Methods:

• Clustering of 311 city microbiome samples using t-SNE.
• Development of a random forest classifier for sample origin prediction.
• Application of de-novo assembly to recover genomic fragments.

Main Results:

• Unsupervised clustering revealed city-specific microbial profiles.
• Geographically close cities (e.g., Auckland, Hamilton) showed similar profiles.
• A random forest classifier achieved 83.3% accuracy in classifying mystery samples.
• Bacterial species like Campylobacter jejuni and Staphylococcus argenteus were highly predictive of origin.
• Sample surface type had minimal impact on microbial profiles.

Conclusions:

• Urban microbiome profiles can be differentiated, aiding city identification.
• Current databases are insufficient for comprehensive urban microbiome characterization.
• Further research is needed to explore the link between geographical distance and microbiome composition.
• The MetaSUB consortium is positioned to inform urban planning and public health policy.