Stability-based comparison of class discovery methods for DNA copy number profiles

Area of Science:

• Genomics
• Bioinformatics
• Cancer Research

Background:

• Array comparative genomic hybridization (array-CGH) analyzes DNA copy number, crucial for tumor development.
• Identifying tumor subclasses with similar array-CGH profiles enhances cancer understanding and treatment.
• Selecting optimal class discovery methods (data representation, dissimilarity, clustering) for array-CGH is challenging.

Purpose of the Study:

• To evaluate and compare different class discovery methods for array-CGH data.
• To identify the most robust combination of methods for classifying array-CGH profiles.
• To address the lack of stability-based comparisons for array-CGH class discovery.

Main Methods:

• Applied multiple class discovery methods to array-CGH data.
• Evaluated method stability using a modified Bertoni's -based test, relaxing independency assumptions.
• Utilized Minimal Regions of alteration for input data representation, sim or agree for dissimilarity measures, and average group distance for clustering.

Main Results:

• The combination of Minimal Regions of alteration, sim/agree dissimilarity, and average group distance clustering produced the most robust classes.
• The modified -based test demonstrated effectiveness in assessing the stability of class discovery solutions.
• Identified specific parameters leading to reliable classification of array-CGH profiles.

Conclusions:

• The optimal class discovery strategy for array-CGH involves specific choices in data representation, dissimilarity measure, and clustering.
• This approach enhances the reliability of identifying cancer subtypes based on genomic profiles.
• The findings provide a framework for selecting robust methods in cancer genomics research.