Identifying joint biomarker panel from multiple level dataset by an optimization model.

Aim: Joint biomarker panel takes advantage of coherence across multiple level datasets and holds the promise to improve disease diagnosis accuracy.

Methods: We collected 101 colorectal cancer and 95 benign samples, measured the molecular concentrations by serum assays and mass spectra, and developed LPGLO (Linear Programming based on Group Lasso Optimization) to identify the joint biomarker panel.

Results: A joint biomarker panel, with ten serum biomarkers and six mass spectra peaks, achieves LOOCV accuracy 0.8724, which is better than the optimal panels identified from separate datasets (LOOCV = 0.7551 for mass spectra only or 0.8265 for serum assay only) and the simply merged dataset (LOOCV = 0.8622).

Conclusion: LPGLO is useful to identify joint biomarker panel to help tumor diagnosis.