Automated sparse feature selection in high-dimensional proteomics data via 1-bit compressed sensing and K-Medoids...

Area of Science:

• Proteomics
• Biomarker Discovery
• Computational Biology

Background:

• High-dimensional proteomics data present challenges in biomarker discovery, including noise, redundancy, and multicollinearity.
• Existing feature selection methods lack stability, sparsity, and computational efficiency for complex proteomic datasets.
• Manual thresholding in conventional methods can lead to suboptimal biomarker identification.

Purpose of the Study:

• To introduce Soft-Thresholded Compressed Sensing (ST-CS), a hybrid framework for automated feature selection in high-dimensional proteomics.
• To address the limitations of current methods in stability, sparsity, and computational efficiency.
• To dynamically partition coefficient magnitudes for accurate discrimination between biomarkers and noise.

Main Methods:

• Integration of 1-bit compressed sensing with K-Medoids clustering.
• Automated feature selection through dynamic partitioning of coefficient magnitudes.
• Dynamic thresholding based on coefficient magnitude distribution.

Main Results:

• ST-CS demonstrated superior feature selection robustness with high sensitivity and specificity, reducing false discovery rates by 20-50% compared to Hard-Thresholded Compressed Sensing (HT-CS).
• Achieved superior F1 scores and Matthews Correlation Coefficients (MCC), outperforming HT-CS, LASSO, and SPLSDA in identifying true biomarkers.
• Outperformed all compared methods in classification accuracy (AUC) across varying noise levels while maintaining sparsity, with significant reductions in selected features on CPTAC datasets.

Conclusions:

• ST-CS rigorously automates feature selection in high-dimensional proteomics.
• The framework effectively balances classification efficacy, interpretability, and scalability for translational proteomics.
• ST-CS offers a powerful and efficient tool for precise biomarker discovery and predictive modeling.