A novel mixed integer programming for multi-biomarker panel identification by distinguishing malignant from benign...

Area of Science:

• Biomedical diagnostics
• Computational biology
• Cancer research

Background:

• Multi-biomarker panels are crucial for complex disease diagnosis, but identifying them is challenging.
• Exhaustive search methods are computationally infeasible for high-dimensional data.

Purpose of the Study:

• To develop a novel method, MILP_k, for identifying serum-based multi-biomarker panels to distinguish colorectal cancer (CRC) from benign tumors.
• To model biomarker panel detection as a mixed integer programming problem to maximize classification accuracy.

Main Methods:

• Proposed MILP_k method using mixed integer programming to identify optimal biomarker panels.
• Analyzed serum profiling data from 101 CRC patients and 95 benign patients.
• Evaluated panel performance using leave-one-out cross-validation (LOOCV) and independent testing with Support Vector Machine (SVM).

Main Results:

• Identified a 4-biomarker panel (CEA, IL-10, IMA, NSE) with 0.7857 LOOCV accuracy and 0.8438 AUC.
• Achieved 20% improvement in predictive accuracy compared to the single best biomarker.
• MILP_k reduced searching time by 1000-fold compared to exhaustive search and identified synergistic biomarker combinations.

Conclusions:

• The MILP_k method efficiently identifies optimal multi-biomarker panels for CRC detection.
• The proposed 4-biomarker panel shows promising diagnostic accuracy and potential clinical utility.
• The method offers a valuable tool for complex disease studies and improves clinical data interpretability.