STENSL: Microbial Source Tracking with ENvironment SeLection.

Area of Science:

• Microbiology
• Bioinformatics
• Machine Learning

Background:

• Microbial source tracking (MST) characterizes microbial communities but is limited to study-specific environments.
• Exploring large public repositories like the Earth Microbiome Project requires methods to select relevant sources from numerous possibilities.

Purpose of the Study:

• To develop a novel machine learning method, STENSL (microbial Source Tracking with ENvironment SeLection), for unsupervised source selection in MST.
• To enable the exploration of latent source environments within large microbial databases.

Main Methods:

• STENSL employs an expectation-maximization algorithm with sparsity to identify contributing microbial environments.
• The method performs unsupervised source selection, differentiating between true sources and nuisance environments.

Main Results:

• STENSL significantly improves the accuracy of true source contribution identification.
• The method effectively reduces noise from noncontributing environments, even with hundreds of potential sources.
• STENSL demonstrates higher accuracy than state-of-the-art methods in complex scenarios.

Conclusions:

• STENSL expands MST capabilities, allowing for the exploration of multiple source environments from public repositories.
• This approach maintains high statistical inference accuracy while enabling automated source exploration and selection.
• STENSL is crucial for leveraging the growing volume of microbiome data for source attribution.