Assessing the statistical validity of proteomics based biomarkers.

Area of Science:

• Proteomics
• Metabolomics
• Biostatistics

Background:

• Biomarker discovery in proteomics and metabolomics often faces challenges with low sample sizes relative to the number of variables (undersampling).
• Existing statistical validation methods may not be sufficient for reliable biomarker identification in such scenarios.

Purpose of the Study:

• To present a comprehensive statistical validation strategy for discrimination models in omics studies.
• To establish a reliable basis for biomarker discovery prior to biochemical validation.

Main Methods:

• Integration of permutation tests, single and double cross-validation.
• Incorporation of a novel variable selection method, rank products.
• Application of principal component discriminant analysis (or any classifier) for classification.

Main Results:

• The proposed strategy demonstrates high performance in an undersampled dataset (Gaucher patients vs. controls).
• Double cross-validation achieved 89% sensitivity and 90% specificity.
• Permutation tests confirmed the suitability of double cross-validation for error rate determination.

Conclusions:

• The combined approach of permutation tests and double cross-validation provides a robust method for validating discrimination models in omics research.
• This strategy effectively mitigates erroneous results common in undersampled datasets, improving biomarker discovery accuracy.