Fuzzy ensemble clustering based on random projections for DNA microarray data analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Unsupervised analysis of gene expression data faces challenges in cluster accuracy, reliability, and defining clear boundaries between patient or gene classes.
• Biological data often exhibits inherent fuzziness, making precise classification difficult.

Purpose of the Study:

• To develop novel clustering methods for improved accuracy and robustness in gene expression data analysis.
• To address the uncertainty in assigning samples to clusters within gene expression datasets.
• To explore strategies that account for the inherent fuzziness of biological data.

Main Methods:

• Proposed a fuzzy ensemble clustering approach utilizing random projections to reduce high-dimensional gene expression data.
• Applied a double fuzzy strategy: fuzzy k-means on projected data, followed by fuzzy t-norm for consensus clustering.
• Investigated variants of fuzzy ensemble algorithms based on different combination techniques.

Main Results:

• Fuzzy ensemble methods applied to leukemia, lymphoma, adenocarcinoma, and melanoma gene expression data showed improved patient class discovery.
• Random projections effectively handled high-dimensional gene expression data, enhancing performance over single fuzzy k-means and resampling-based ensembles.
• Analysis of base fuzzy clustering accuracy and diversity provided insights into the ensemble approach's strengths and weaknesses.

Conclusions:

• The proposed fuzzy ensemble clustering approach can improve the discovery of bio-molecular patient classes by embracing data fuzziness.
• Dimensionality reduction via random projections is suitable for high-dimensional gene expression data, leading to better performance.
• Understanding clustering accuracy and diversity is key to leveraging fuzzy ensemble methods effectively.