Latent Model-Based Clustering for Biological Discovery.

Area of Science:

• Computational biology
• Bioinformatics
• Data science

Background:

• Biological data analysis often requires robust clustering methods.
• Existing methods may struggle with overlapping clusters or scalability.
• Latent model-based approaches offer a promising avenue for complex biological datasets.

Purpose of the Study:

• To introduce LOVE (Latent model-based Overlapping Variable-based Ensemble) as a novel clustering method.
• To demonstrate LOVE's capability in generating both overlapping and non-overlapping clusters.
• To validate LOVE's performance and biological relevance across diverse datasets.

Main Methods:

• LOVE utilizes a latent model and an allocation matrix to define clusters.
• The method ensures unique definition of the allocation matrix and corresponding clusters.
• LOVE was applied to gene expression, HIV serological responses, and vaccine-induced immune response datasets.

Main Results:

• LOVE successfully generated meaningful biological clusters across all tested datasets.
• Clusters identified by LOVE exhibited stability across various tuning parameters.
• LOVE demonstrated superior performance compared to 13 state-of-the-art clustering methods.

Conclusions:

• LOVE is a robust and scalable method for biological discovery through clustering.
• The method provides accurate and meaningful overlapping and non-overlapping clusters.
• LOVE has broad applicability across large-scale biological datasets.