Consensus clustering for Bayesian mixture models.

Area of Science:

• Computational biology
• Bioinformatics
• Data science

Background:

• Cluster analysis is crucial for precision medicine and systems biology.
• Consensus clustering, an ensemble method, improves robustness by combining multiple clustering runs.
• Existing methods face challenges with computational scalability and complex datasets.

Purpose of the Study:

• To develop a computationally scalable and robust consensus clustering approach.
• To adapt Bayesian mixture models for heuristic clustering with early stopping criteria.
• To enable effective clustering analysis for large and complex biological datasets.

Main Methods:

• Applied consensus clustering to heuristic algorithms derived from Bayesian mixture models with early stopping.
• Developed heuristics for determining ensemble size and early stopping criteria.
• Integrated the approach with a Bayesian integrative clustering method for multi-omics data analysis.

Main Results:

• Successfully uncovered target clustering structures in simulation studies.
• Demonstrated significant runtime reductions compared to traditional Bayesian inference, especially in parallel environments.
• Identified co-expressed gene clusters with shared regulatory proteins in budding yeast multi-omics data.

Conclusions:

• The proposed method acts as a wrapper for existing Bayesian clustering tools, extending their applicability.
• Enables meaningful clustering analyses even with computational limitations or poor model convergence.
• Facilitates the analysis of larger datasets and sophisticated models, including multi-dataset integration.