Clustering on Human Microbiome Sequencing Data: A Distance-Based Unsupervised Learning Model.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Human microbiome analysis is crucial for understanding health and disease.
• Microbiome sequencing data (16S rRNA) is often sparse, skewed, and contains excess zeros.
• Standardized clustering methods are lacking for complex microbiome data.

Purpose of the Study:

• To develop a novel clustering algorithm for microbiome data.
• To address the presence-absence bias in sparse count data.
• To improve patient stratification in personalized medicine.

Main Methods:

• Proposed a clustering algorithm utilizing a novel beta diversity measure.
• Developed a distance measure derived from a parametric mixture model.
• Estimated sample-specific distributions and mixture weights for OTU counts.

Main Results:

• The proposed method accurately estimates true zero proportions.
• The new beta diversity measure precisely quantifies sample distances.
• Achieved substantial clustering improvement over existing methods on simulated sparse data.

Conclusions:

• The novel clustering approach effectively handles zero-inflated microbiome data.
• The method enhances sample stratification for personalized medicine applications.
• Successfully applied to identify distinct microbiome states in Parkinson's disease.