Rank-based learning: a novel high-throughput algorithm resilient to missing data and effective for datasets with...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• High-throughput omics data present classification challenges due to platform variability, batch effects, missing values, and high dimensionality.
• Existing methods struggle with noise and inconsistencies inherent in omics data, limiting diagnostic model reliability.

Purpose of the Study:

• To introduce and evaluate a novel Rank-Based Learning (RBL) method for binary classification of high-throughput omics data.
• To enhance the robustness and generalizability of diagnostic models by leveraging relative feature rankings.

Main Methods:

• Developed a Rank-Based Learning (RBL) algorithm focusing on relative feature rankings.
• Evaluated RBL against Logistic Regression (LR) and Random Forest (RF) using simulated data.
• Validated RBL on two real-world plasma proteomics datasets: small cell lung cancer (SCLC) and duodenopancreatic neuroendocrine tumors (dpNET) in MEN1 patients.

Main Results:

• RBL outperformed LR and RF in simulation experiments, particularly under batch effects and missing data conditions.
• In SCLC classification, RBL achieved a test AUC of 0.76, superior to LR (0.65) and RF (0.59).
• For dpNET, RBL demonstrated strong performance with an AUC of 0.80 on the test set, outperforming LR (0.57) and RF (0.53).

Conclusions:

• Rank-Based Learning (RBL) effectively mitigates non-biological variation by emphasizing feature rankings over absolute expression levels.
• RBL significantly improves predictive accuracy for diagnostic models using complex omics data.
• The RBL framework offers a promising avenue for developing more reliable and clinically applicable omics-based diagnostic tools.