BC_Intrinsic_subtyping

BC_Intrinsic_subtyping assigns breast cancer (BC) RNA-sequencing (RNA-seq) expression profiles to intrinsic molecular subtypes to support stratification and prognostic assessment.

Key Features:

• Adapted PAM50 classifier: Incorporates an improved dataset-level PAM50 classifier originally developed for microarray data, adapted to classify BC into five intrinsic subtypes from RNA-seq.
• AWCA (Adaptive Weighting of Centroids Algorithm): Applies AWCA to mitigate cohort-composition effects on PAM50, improving concordance with existing classifications to over 90% while preserving prognostic signal.
• Single-sample methodology: Enables AWCA-enhanced PAM50 classification on individual samples without reference cohorts.
• Machine learning integration: Evaluates supervised learners and identifies regularized multiclass logistic regression (mLR) as the most effective single-sample classifier with high concordance to PAM50 on external test sets.
• Gene selection optimization: Uses ad-hoc gene selection from the global transcriptome to enhance mLR classification accuracy and prognostic performance without reference samples.

Scientific Applications:

• Clinical utility: Supports molecular subtype–based patient stratification for treatment planning and prognosis assessment in breast cancer.
• Research enhancement: Enables robust intrinsic subtype analysis across RNA-seq datasets for large-scale genomic studies and validation of transcriptional classifications.
• Methodological advancement: Provides AWCA and ML-based single-sample classifiers to improve reliability and portability of intrinsic subtyping from RNA-seq.

Methodology:

Uses an adapted dataset-level PAM50 classifier for RNA-seq, the AWCA (Adaptive Weighting of Centroids Algorithm), supervised machine learning including regularized multiclass logistic regression (mLR), and ad-hoc gene selection from the global transcriptome.