Assessment of hierarchical clustering methodologies for proteomic data mining

Area of Science:

• Proteomics
• Bioinformatics
• Data Mining

Background:

• Hierarchical clustering is a valuable tool for initial exploration of proteomic data, grouping samples or proteins based on expression profiles.
• Clustering outcomes are sensitive to parameter choices, including data preprocessing, similarity measures, and dendrogram construction.
• Evaluating different clustering strategies is crucial for optimizing the analysis of complex biological datasets.

Purpose of the Study:

• To assess and identify optimal hierarchical clustering strategies for proteomic data analysis.
• To compare the performance of various parameter combinations using a robust quality metric.
• To provide guidance on selecting effective clustering methods for proteomic expression profiling.

Main Methods:

• Utilized hierarchical clustering methodology for proteomic data exploration.
• Employed the F-measure as a primary metric for evaluating clustering quality.
• Investigated various parameter combinations, including data preprocessing (log transformation), similarity measures (Pearson correlation), and aggregation methods (Ward's method).

Main Results:

• The combination of log-transformed data, Pearson correlation for similarity, and Ward's method for aggregation emerged as a highly effective clustering strategy.
• This specific strategy demonstrated superior performance across the studied proteomic datasets.
• PermutMatrix software, adapted from transcriptomics, was used to facilitate these analyses.

Conclusions:

• The optimal hierarchical clustering strategy for the analyzed proteomic datasets involves using log-transformed data, Pearson correlation, and Ward's method.
• This validated approach enhances the reliability of grouping samples and proteins based on expression patterns.
• The findings offer practical recommendations for researchers utilizing clustering in proteomic data mining.