Disentangling Prognostic and Predictive Biomarkers Through Mutual Information

Area of Science:

• Biostatistics
• Information Theory
• Biomarker Discovery

Background:

• Distinguishing between predictive and prognostic biomarkers is crucial in clinical trials.
• Existing methods for biomarker ranking present challenges in disentangling their distinct roles.

Purpose of the Study:

• To develop and apply information theoretic methods for ranking biomarkers.
• To formalize biomarker ranking as an optimization problem using information theoretic quantities.
• To introduce a novel visualization tool for assessing biomarker predictive and prognostic strengths.

Main Methods:

• Formulating biomarker ranking via optimization of information theoretic quantities.
• Estimating high-dimensional conditional mutual information terms for biomarker assessment.
• Employing efficient low-dimensional approximations for term estimation.

Main Results:

• Derivation of rankings for predictive and prognostic biomarkers.
• Successful estimation of complex information theoretic terms.
• Development of a visualization tool to represent biomarker prognostic and predictive power.

Conclusions:

• The proposed information theoretic framework offers a robust method for biomarker ranking.
• The new visualization tool aids in understanding and discovering biomarker utility.
• This approach is expected to significantly advance biomarker discovery in clinical settings.