Finding disagreement pathway signatures and constructing an ensemble model for cancer classification

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-throughput molecular profiling enables cancer classification but faces challenges with large gene numbers relative to sample size.
• Current gene selection methods often neglect biological principles, limiting their effectiveness with biological data.

Purpose of the Study:

• To develop a robust ensemble learning paradigm incorporating multiple pathways information for improved cancer classification.
• To compare the proposed method against existing techniques like Elastic SCAD and PPDMF.
• To investigate the biological mechanisms and clinical relevance of selected genes and pathways.

Main Methods:

• Proposed a novel ensemble learning framework integrating multi-pathway information for gene selection.
• Compared classification performance using various metrics against established statistical and machine learning methods.
• Conducted functional annotation and pathway analysis on identified feature genes.

Main Results:

• The proposed ensemble method demonstrated superior performance across most metrics compared to other approaches.
• Identified ensemble feature genes significantly associated with drug targets and clinically relevant cancer.
• Discovered core biological pathways and processes linked to clinically relevant phenotypes.

Conclusions:

• The developed ensemble learning paradigm offers a robust and biologically informed approach to cancer classification.
• Selected genes and pathways provide insights into cancer mechanisms and potential therapeutic targets.
• This research offers a new perspective for studying molecular activities in cancer.