A comprehensive evaluation of module detection methods for gene expression data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Analyzing large genome-wide gene expression datasets requires effective module detection.
• Classical clustering methods have limitations for co-expression module identification.
• Alternative methods address limitations by using sample subsets, regulatory networks, or overlapping modules.

Purpose of the Study:

• To comprehensively evaluate and compare various gene co-expression module detection methods.
• To assess the performance of different strategies using known regulatory networks.
• To provide recommendations for improving module detection techniques.

Main Methods:

• Utilized known regulatory networks for a robust evaluation of module detection methods.
• Compared classical clustering with alternative approaches like decomposition, biclustering, and network inference.
• Investigated practical aspects including parameter estimation and similarity measures.

Main Results:

• Decomposition methods demonstrated superior performance over other strategies.
• Biclustering and network inference approaches did not show a clear advantage on large datasets.
• The evaluation workflow identified key factors in module detection efficacy.

Conclusions:

• Decomposition methods are recommended for co-expression module detection in large gene expression datasets.
• Further research should focus on refining decomposition methods and exploring their practical implementation.
• The study provides a framework for evaluating and developing advanced module detection tools.