Machine Learning-Based Ensemble Recursive Feature Selection of Circulating miRNAs for Cancer Tumor Classification

Area of Science:

• Biochemistry
• Bioinformatics
• Oncology

Background:

• Circulating microRNAs (miRNAs) are noncoding RNA molecules found in bodily fluids, offering a minimally invasive approach for cancer detection and subtyping.
• Machine learning has been applied to miRNA datasets for tumor classification, but results are often complex and difficult for medical experts to interpret.
• Current methods analyze thousands of miRNAs, hindering clinical application and understanding.

Purpose of the Study:

• To develop a novel technique for reducing the number of circulating miRNAs needed for accurate tumor classification.
• To create a more interpretable and clinically actionable precision medicine pipeline using miRNA biomarkers.
• To demonstrate the feasibility of dimensionality reduction for circulating miRNA-based diagnostics.

Main Methods:

• A recursive feature elimination procedure was employed, integrating a heterogeneous ensemble of state-of-the-art classifiers.
• The approach utilizes ensemble methods to mitigate classifier biases and feature selection to address data batch effects.
• The methodology was tested on a 10-cancer type classification task and a specific breast cancer subtype classification task.

Main Results:

• The proposed technique effectively reduced the set of informative circulating miRNAs for classification tasks.
• The method demonstrated robust and reliable outcomes in both general cancer type and specific subtype classification.
• Performance favorably compared to existing state-of-the-art feature selection methods.

Conclusions:

• Dimensionality reduction using a heterogeneous ensemble with recursive feature elimination is effective for circulating miRNA-based cancer classification.
• This approach simplifies complex miRNA data, making it more interpretable for clinical applications.
• The methodology represents a significant step toward clinically actionable, miRNA-driven precision oncology.