Biomarker clustering to address correlations in proteomic data.

Area of Science:

• Proteomics
• Bioinformatics
• Statistical Analysis

Background:

• Correlated variables confound statistical analyses in microarray and proteomics experiments.
• Biological and technical effects in mass spectrometry (MS) exacerbate data correlation.
• Existing dimension reduction methods lack clear biological interpretation.

Purpose of the Study:

• To propose a novel preprocessing algorithm for handling correlated features in MS data.
• To improve statistical power and biomarker discovery by analyzing clusters instead of individual features.
• To address limitations of existing dimension reduction techniques.

Main Methods:

• Developed a preprocessing algorithm that clusters highly correlated features.
• Utilized the Bayes information criterion to determine the optimal number of clusters.
• Applied statistical analysis to clusters, reducing noise and mitigating correlation issues.

Main Results:

• The clustering approach demonstrated increased statistical power using false discovery rate on simulated data.
• Analysis of real-world SELDI-TOF-MS datasets showed improved biomarker discovery.
• Successfully applied to clinical datasets for Kawasaki disease and leukemia.

Conclusions:

• Clustering highly correlated features is an effective preprocessing step for MS data.
• This method enhances statistical power and aids in biomarker discovery.
• The algorithm shows promise for clinical proteomics applications.