Microbiome Preprocessing Machine Learning Pipeline.

Area of Science:

• Microbiome analysis
• Bioinformatics
• Machine Learning

Background:

• 16S gene sequencing data is commonly utilized for machine learning (ML) tasks.
• Feature counts from 16S sequences represent taxonomic information but may not be optimal for ML.
• Raw feature counts can be suboptimal for ML model training and prediction.

Purpose of the Study:

• To identify and evaluate optimal preprocessing steps for 16S sequencing data in ML classification.
• To determine the impact of various preprocessing techniques on classification accuracy.
• To develop and integrate effective preprocessing strategies for microbiome ML applications.

Main Methods:

• Systematic evaluation of multiple preprocessing steps for 16S sequencing data.
• Calculation of Area Under the Curve (AUC) to quantify classification accuracy improvements.
• Testing combinations of preprocessing techniques including log transformation, feature merging, and z-scoring.

Main Results:

• Log-transformed feature counts demonstrate higher informativeness compared to relative counts.
• Merging features by taxonomic level using dimension reduction enhances AUC.
• Z-scoring preprocessing yields a negligible effect on classification results.

Conclusions:

• Effective preprocessing of microbiome 16S data is critical for achieving optimal ML performance.
• The developed Microbiome Preprocessing Machine Learning Pipeline (MIPMLP) integrates these optimized steps.
• MIPMLP is available as a standalone tool and a web service for public use.