A resample-replace lasso procedure for combining high-dimensional markers with limit of detection

Area of Science:

• Biostatistics
• Biomarker Discovery
• High-Throughput Technologies

Background:

• Biomarker combinations enhance disease screening sensitivity over individual markers.
• High-dimensional data from high-throughput technologies pose challenges for traditional parametric methods due to complex covariance matrices.
• Instrumental limit of detection (LOD) creates missing data (NA values), complicating biomarker analysis.

Purpose of the Study:

• To develop a robust method for combining high-dimensional biomarkers in the presence of missing data due to LOD.
• To address the computational challenges of inverting covariance matrices for high-dimensional datasets.
• To improve the accuracy of disease diagnosis through effective biomarker combination.

Main Methods:

• Proposed a resample-replace lasso procedure for biomarker combination.
• Imputed biomarker values below the LOD.
• Employed the graphical lasso method to estimate means and precision matrices for high-dimensional biomarkers.

Main Results:

• The resample-replace lasso procedure outperformed alternative methods (e.g., substituting NA with LOD, removing NA values).
• Simulation studies validated the proposed method's effectiveness.
• Real-case analysis demonstrated superior accuracy in distinguishing glioblastoma patient survival groups.

Conclusions:

• The resample-replace lasso procedure offers a superior approach for high-dimensional biomarker combination with LOD data.
• This method enhances diagnostic accuracy in complex biological studies.
• The technique is particularly valuable for analyzing protein profiling data in cancer research.