A Stack-based Ensemble Framework for Detecting Cancer MicroRNA Biomarkers.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial regulators of gene expression with roles in biological processes.
• Aberrant miRNA expression profiles are linked to oncogenesis, showing differential patterns in normal versus tumor tissues.
• Automatic classification of miRNAs based on expression similarity for cancer research remains an underexplored area.

Purpose of the Study:

• To propose a novel framework for the automatic classification of cancer, miRNA, and mRNA expression datasets.
• To address the challenge of classifying miRNAs by considering expression value similarities.
• To improve the accuracy of cancer subtyping and biomarker identification using expression data.

Main Methods:

• A two-stage approach combining multiobjective optimization and ensemble learning.
• Stage 1: Non-dominated Sorting Genetic Algorithm II (NSGA-II) for automatic selection of classifier type, parameters, and features.
• Stage 2: Stack-based ensemble technique to integrate solutions from Stage 1 for a robust combinatorial classification.

Main Results:

• The proposed two-stage framework demonstrated superior classification accuracy on various cancer and RNA expression profile datasets.
• The method outperformed several state-of-the-art approaches in classifying different datasets.
• Effective identification of relevant features and classifier parameters for improved diagnostic accuracy.

Conclusions:

• The developed two-stage framework offers a powerful and accurate solution for classifying complex biological expression datasets.
• This approach holds significant potential for advancing cancer research and diagnostics through improved miRNA and mRNA expression analysis.
• The integration of multiobjective optimization and ensemble methods provides a robust strategy for biological data classification.