Supervised Machine Learning Enables Geospatial Microbial Provenance.

Area of Science:

• Microbiology
• Bioinformatics
• Machine Learning

Background:

• Geospatial metadata in metagenomic datasets enables the identification of location-specific microbial signatures.
• Urban microbiomes are increasingly studied for environmental and forensic applications.

Purpose of the Study:

• To determine the regional specificity of environmental metagenomes.
• To develop and evaluate machine learning methods for identifying microbial fingerprints.
• To create a novel classifier for bioindicator species and microbial interactions.

Main Methods:

• Analysis of 4305 shotgun-sequenced samples from the MetaSUB Consortium (urban microbiomes).
• Application and comparison of ten supervised machine learning (SML) classifiers.
• Development and testing of the Cassandra random-forest-based classifier.
• Validation on the Tara Oceans dataset (marine microbiomes).

Main Results:

• City-specific microbial fingerprints were identified with high accuracy using SML (85-89% for cities, 90-94% for continents).
• The Cassandra algorithm achieved 83% accuracy in classifying oceanic sample locations.
• Cassandra can identify bioindicator species and infer microbial interactions.

Conclusions:

• SML methods and the Cassandra classifier are effective for identifying microbial fingerprints in diverse environments.
• This approach supports biotracing, environmental monitoring, and microbial forensics.
• The study highlights the utility of large-scale microbiome datasets for regional specificity analysis.