Partial mixture model for tight clustering of gene expression time-course

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Tight clustering aims for more informative gene expression clusters by excluding loosely correlated genes.
• Maximum likelihood techniques are common for parameter estimation, while minimum distance estimators are underutilized.
• Limited research exists on robust methods for excluding scattered genes in clustering.

Purpose of the Study:

• To develop a robust minimum distance estimator for parameter estimation in model-based time-course clustering.
• To formulate a partial mixture model that accommodates replicate information and scattered genes.
• To evaluate the performance of the proposed clustering algorithm against existing methods.

Main Methods:

• Formulation of a partial mixture model for time-course gene expression data.
• Application of the minimum distance estimator for parameter estimation.
• Development of a partial regression clustering algorithm.
• Validation using simulated and real gene expression datasets.

Main Results:

• The minimum distance estimator demonstrates inherent robustness and superior performance compared to the maximum likelihood estimator in simulated data.
• The proposed partial regression clustering algorithm achieved top scores in Gene Ontology-driven evaluation against four other algorithms.
• Biological and statistical validations confirmed the algorithm's effectiveness on diverse datasets.

Conclusions:

• The partial mixture model is successfully extended for time-course data analysis.
• The combination of the partial mixture model and minimum distance estimator is suitable for tight clustering.
• The algorithm provides deeper biological understanding and generates new hypotheses, highlighting the relevance of scattered genes.