Identification of prognostic and predictive biomarkers in high-dimensional data with PPLasso

Area of Science:

• Biostatistics
• Genomics
• Precision Medicine

Background:

• Biomarker identification is crucial for precision medicine.
• Genomic data offers vast potential but presents challenges due to high dimensionality and correlations.
• Existing methods often overlook the simultaneous selection of prognostic and predictive biomarkers.

Purpose of the Study:

• To develop a novel statistical method for simultaneously selecting prognostic and predictive biomarkers from high-dimensional genomic data.
• To address the challenge of highly correlated biomarkers in biomarker selection.
• To improve the accuracy of biomarker identification for clinical applications.

Main Methods:

• Propose PPLasso, a novel approach integrating prognostic and predictive effects into a single statistical model.
• Transform the design matrix to remove correlations between biomarkers prior to applying generalized Lasso.
• Evaluate PPLasso's performance against traditional Lasso and other extensions.

Main Results:

• PPLasso demonstrates superior performance in identifying both prognostic and predictive biomarkers compared to traditional methods.
• The method effectively handles high-dimensional and correlated genomic data.
• Successful application to publicly available transcriptomic and proteomic datasets.

Conclusions:

• PPLasso offers a robust and accurate approach for simultaneous prognostic and predictive biomarker selection.
• The method enhances the utility of genomic data in precision medicine.
• PPLasso provides a valuable tool for clinical trial design and patient stratification.