Differential protein expression and peak selection in mass spectrometry data by binary discriminant analysis.

Area of Science:

• Biomedical data analysis
• Proteomics
• Clinical diagnostics

Background:

• Proteomic mass spectrometry is crucial for clinical diagnostics, particularly for monitoring cancer biomarkers in blood.
• Identifying differentially expressed proteins and relevant peaks for class separation in proteomics remains a significant challenge.

Purpose of the Study:

• To introduce a novel, effective approach for identifying differentially expressed proteins using binary discriminant analysis.
• To provide a computationally inexpensive method for analyzing mass spectrometry data in both two-group and multi-group settings.

Main Methods:

• Developed a data-adaptive thresholding approach for protein expression values.
• Employed relative entropy for ranking dichotomized features.
• Generalized the 'peak probability contrast' method.

Main Results:

• The binary discriminant analysis approach achieved prediction accuracy equivalent to random forest on a large-scale drug discovery dataset.
• Identified statistically predictive and biologically relevant marker peaks in pancreas cancer mass spectrometry data, which were previously unrecognized.
• The method is computationally inexpensive and applicable to both two-group and multi-group comparisons.

Conclusions:

• Binary discriminant analysis offers a powerful and efficient tool for differential proteomics.
• This approach enhances the identification of clinically relevant biomarkers from mass spectrometry data.
• The R package 'binda' is available for implementing this methodology.