Machine learning clustering and classification of human microbiome source body sites

Area of Science:

• Microbiology
• Forensic Science
• Bioinformatics

Background:

• Human body sites harbor distinct microbial communities.
• Current methods for identifying biological material origins from crime scenes are limited.
• Metagenomics and machine learning offer potential advancements.

Purpose of the Study:

• To investigate the utility of machine learning (ML) for forensic source body site identification.
• To determine if microbiomic signatures can accurately discriminate between human body sites.
• To assess the predictive power of ML algorithms on metagenomic data.

Main Methods:

• Utilized shotgun metagenomic sequencing data.
• Applied machine learning clustering algorithms for taxonomic classification.
• Analyzed microbial signatures at genus, family, and order levels.

Main Results:

• Achieved over 99% purity in predicting source body site at the genus taxonomy.
• Consistent accuracy was observed across genus, family, and order taxonomies.
• A core set of 51 genera also yielded accurate predictions.

Conclusions:

• Machine learning effectively identifies human body site origins using microbial signatures.
• The developed method offers a replicable and accurate approach for forensic investigations.
• Integration of ML predictors is recommended for forensic source identification protocols.