RNA-Seq-Based Breast Cancer Subtypes Classification Using Machine Learning Approaches.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Breast invasive carcinoma (BRCA) comprises distinct subtypes, each with unique morphology.
• Understanding gene interaction mechanisms is crucial for personalized BRCA treatment.
• Current computational methods for BRCA subtype identification have limitations in fully elucidating gene interactions.

Purpose of the Study:

• To identify and explore gene interaction mechanisms specific to each BRCA subtype.
• To develop a computational approach for accurate BRCA subtype classification.
• To provide insights into biological function changes across BRCA subtypes.

Main Methods:

• Integrated gene regulatory networks and differential expression analysis to obtain weighted differentially expressed genes (weighted DEGs).
• Constructed gene coexpression networks for control and experiment groups.
• Developed Gene Ontology (GO) enrichment based on gene coexpression networks (GOEGCN) for two-side distinction analysis.

Main Results:

• Modeled binary classification using weighted DEGs, achieving good prediction performance for unseen samples.
• Validated the effectiveness of the proposed weighted DEG and GOEGCN approaches.
• Identified novel enriched GO terms via GOEGCN, explaining specific biological function changes in BRCA subtypes.

Conclusions:

• Weighted DEGs derived from gene regulatory networks hold significant biological importance for BRCA subtypes.
• Five binary classifiers based on weighted DEGs demonstrated strong performance across key metrics (Sensitivity, Specificity, Accuracy, F1, AUC).
• The GOEGCN method provides novel insights into biological function alterations in BRCA subtypes through enriched GO terms.