SMART: unique splitting-while-merging framework for gene clustering.

Area of Science:

• Computer Science
• Data Mining
• Bioinformatics

Background:

• Clustering algorithm performance heavily relies on parameter settings.
• Difficulty in setting parameters a priori leads to degraded clustering performance.
• Existing self-splitting algorithms often over-cluster and then merge, lacking dynamic adaptation.

Purpose of the Study:

• Propose a novel clustering framework, splitting merging awareness tactics (SMART).
• Develop a framework that eliminates the need for a priori knowledge of cluster number or range.
• Integrate multiple clustering techniques for robust and reliable results.

Main Methods:

• Implemented SMART with competitive learning and finite mixture models.
• Integrated the minimum message length algorithm as a clustering selection criterion.
• Tested the framework on demonstration, simulated gene expression, and real microarray gene expression datasets.

Main Results:

• SMART demonstrated superior performance compared to existing self-splitting and traditional algorithms across various metrics.
• Numerical results confirmed the effectiveness of the SMART framework and its derived algorithms.
• The framework successfully handled real microarray gene expression data analysis.

Conclusions:

• SMART requires no dataset-dependent parameters or a priori knowledge.
• The framework is extendible to diverse applications beyond current implementations.
• SMART offers superior clustering performance, addressing limitations of current algorithms.