Proximity measures for clustering gene expression microarray data: a validation methodology and a comparative...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Cluster analysis is a primary step for extracting insights from gene expression microarray data.
• Selecting an appropriate proximity measure is critical for effective clustering, yet guidelines are lacking.
• Pearson is widely used, but the performance of other measures remains largely unexamined.

Purpose of the Study:

• To investigate the impact of different proximity measures on microarray data clustering.
• To evaluate the performance of 16 proximity measures across diverse gene expression datasets.
• To provide guidance on selecting proximity measures for specific experimental contexts.

Main Methods:

• Evaluated 16 proximity measures on 52 gene expression microarray datasets from cancer and time-course experiments.
• Developed a benchmark and a novel methodology, Intrinsic Biological Separation Ability (IBSA), for time-course data evaluation.
• Compared the performance of commonly used measures (e.g., Pearson, Spearman, Euclidean distance) against less explored alternatives.

Main Results:

• Measures not frequently used in gene expression literature demonstrated superior performance compared to standard measures.
• The optimal proximity measure varied significantly between time-course and cancer gene expression data.
• The IBSA methodology and benchmark offer a standardized approach for future research on time-course data.

Conclusions:

• The choice of proximity measure for clustering gene expression data should be tailored to the specific experimental scenario (e.g., time-course vs. cancer).
• Rarely used proximity measures can yield better clustering results than commonly employed ones.
• The developed IBSA methodology and benchmark are valuable resources for advancing research in gene time-course data analysis.