Balancing the robustness and predictive performance of biomarkers

Area of Science:

• Biostatistics
• Biomarker Discovery
• Proteomics

Background:

• Assessing biomarker robustness is crucial for reliable diagnostics.
• Stable biomarkers are consistently identified despite minor data variations.
• Stability alone is insufficient; diagnostic utility must also be considered.

Purpose of the Study:

• To develop strategies combining stability and predictive performance for robust biomarker identification.
• To evaluate the impact of biomarker correlations on stability assessments.
• To identify robust candidate biomarkers for HTLV1-associated myelopathy/tropical spastic paraparesis (HAM/TSP).

Main Methods:

• Proposed strategies integrate stability assessment with predictive performance evaluation.
• Logistic regression classifiers with elastic net regularization were employed.
• A synthetic dataset was used for initial strategy illustration.

Main Results:

• Stability assessments can decrease false positives but may miss true positives.
• Combining stability and predictive performance enhances the true positive rate.
• Biomarker correlations negatively affect stability and require careful consideration.

Conclusions:

• The proposed strategies effectively identify robust and diagnostically useful biomarkers.
• Accounting for biomarker correlations is essential for accurate stability assessment.
• Candidate robust biomarkers for HAM/TSP were identified using these methods.