Definition of the microbial rare biosphere through unsupervised machine learning

Area of Science:

• Microbial Ecology
• Bioinformatics
• Computational Biology

Background:

• The microbial rare biosphere, comprising low-abundance microorganisms, lacks standardized definition methods.
• Current reliance on arbitrary abundance thresholds (e.g., 0.1%) hinders cross-study comparisons.
• A robust, objective approach is needed to delineate rare microbial taxa.

Purpose of the Study:

• To develop and present ulrb (Unsupervised Learning based Definition of the Rare Biosphere), an R package for objective rare biosphere delineation.
• To provide a user-independent method for classifying microbial taxa into abundance categories.
• To improve consistency and comparability in rare biosphere studies.

Main Methods:

• Utilized unsupervised machine learning algorithms to classify taxa into abundance categories (rare, intermediate, abundant).
• Developed an R package, ulrb, implementing the unsupervised learning approach.
• Validated the method's consistency against arbitrary threshold-based approaches.

Main Results:

• ulrb offers a more consistent method for defining the microbial rare biosphere compared to arbitrary thresholds.
• The package is applicable to diverse datasets from microbial ecology and other fields.
• ulrb successfully identifies different types of rarity and is statistically valid across various dataset sizes.

Conclusions:

• ulrb provides a user-independent and statistically valid method for discerning rare from abundant organisms.
• This approach enhances the comparability and reliability of microbial rare biosphere research.
• The R package ulrb is a valuable tool for ecological data analysis.