Randomized feature selection based semi-supervised latent Dirichlet allocation for microbiome analysis.

Area of Science:

• Microbiome research
• Computational biology
• Health informatics

Background:

• The gut microbiome significantly influences health and disease.
• Accurate microbiome analysis is vital for improved diagnostics and treatments.
• Self-reported, fuzzy health labels pose challenges for traditional supervised learning.

Purpose of the Study:

• To develop a novel semi-supervised methodology for analyzing gut microbiome data and classifying subject health status.
• To address the limitations of supervised learning with fuzzy health labels.
• To enhance the understanding of the gut microbiome's impact on individual well-being.

Main Methods:

• Employed Latent Dirichlet Allocation (LDA) for unsupervised clustering of microbiome data.
• Integrated observed health status to create a semi-supervised classification approach.
• Incorporated Randomized Feature Selection (RFSLDA) to improve classification performance and handle high-dimensional data.

Main Results:

• The proposed RFSLDA method demonstrated high classification accuracy for health status.
• RFSLDA outperformed popular supervised learning methods like Support Vector Machines (SVM) and multinomial logistic models.
• The framework effectively identified key bacterial types indicative of health status and enhanced subject similarity assessment.

Conclusions:

• RFSLDA offers an effective and efficient semi-supervised topic modeling approach for microbiome association studies.
• The method enhances clustering accuracy by identifying crucial bacteria for health status classification.
• RFSLDA facilitates the identification of key bacterial indicators and highly similar subjects based on health status.