Accurate confidence aware clustering of array CGH tumor profiles

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Chromosomal aberrations are characteristic of cancer subtypes and detectable via array comparative genomic hybridization (aCGH).
• Clustering aCGH tumor profiles aids in identifying key chromosomal regions and provides cancer diagnostic information.
• Reliably assigning individual aCGH tumor profiles to specific cancer type clusters is a critical challenge.

Purpose of the Study:

• To introduce a novel evolutionary fuzzy clustering (EFC) algorithm for analyzing aCGH tumor profiles.
• To assess the algorithm's ability to handle overlapping clusterings and provide confidence measures for sample-to-type assignment.
• To evaluate the performance of EFC against existing methods on real-world cancer datasets.

Main Methods:

• Development of a novel evolutionary fuzzy clustering (EFC) algorithm.
• Utilizing cluster membership degrees as confidence measures for sample assignment.
• Testing EFC on synthetic and four real-world aCGH tumor profile datasets across different cancer types.

Main Results:

• The EFC algorithm effectively handles overlapping clusterings using membership degrees.
• EFC outperforms existing clustering methods on four real aCGH tumor profile datasets.
• The 1-Pearson correlation coefficient is identified as the optimal distance measure, and preprocessing steps like segmentation can decrease performance.

Conclusions:

• The EFC algorithm provides a robust method for classifying cancer types based on aCGH profiles.
• Cluster membership degrees offer valuable confidence measures for diagnostic assignments.
• The study highlights the importance of the chosen distance measure and advises against certain preprocessing steps for optimal clustering performance.