A Comparison of Fuzzy Clustering Approaches for Quantification of Microarray Gene Expression

Area of Science:

• Biomedical Imaging
• Bioinformatics
• Computational Biology

Background:

• Microarray imaging is vital for biomedical research but faces clinical reliability challenges.
• Accurate spot segmentation and gene expression quantification (mRNA) are critical issues.
• Existing software struggles with complex spot shapes like donuts and scratches.

Purpose of the Study:

• To evaluate fuzzy clustering approaches for robust microarray spot segmentation.
• To compare the performance of different fuzzy clustering algorithms.
• To identify improved statistical methods for gene expression quantification.

Main Methods:

• Application of fuzzy clustering techniques, including Possibilistic c-means (PCM), for pixel-level soft labeling.
• Testing segmentation performance on simulated and real microarray images.
• Comparison of three statistical criteria for measuring gene expression levels.

Main Results:

• Possibilistic c-means (PCM) demonstrated superior performance based on stability criteria for spot segmentation.
• A novel asymptotically unbiased statistic provided more accurate gene expression quantification.
• Fuzzy clustering overcomes limitations of hard-labeling methods like k-means for complex spot shapes.

Conclusions:

• Fuzzy clustering, particularly PCM, significantly enhances microarray spot segmentation reliability.
• Improved quantification methods increase the accuracy of gene expression level measurements.
• These advancements pave the way for more dependable clinical applications of microarray imaging.