DeepCC

DeepCC classifies cancer molecular subtypes from gene expression profiles using supervised deep learning on functional spectra that quantify biological pathway activities.

Key Features:

• Deep learning-based framework: Employs supervised deep learning to analyze functional spectra that quantify activities of biological pathways for subtype discrimination.
• Operates on gene expression profiles: Trains models on gene expression data that capture pathway-level functional activities.
• Platform independence: Mitigates platform differences and batch effects that affect gene expression signature-based classifications.
• Robustness to missing data: Demonstrates robustness via simulation analysis using random subsampling of genes to emulate incomplete datasets.
• Single-sample classification: Supports classification of individual patient samples at the single-sample level.
• Deep feature extraction and separation: Extracts deep features that produce more compact within-subtype distributions and clearer separation between subtypes, reducing unclassifiable samples.
• Comparative performance: Validated against random forests (RF), support vector machines (SVM), gradient boosting machines (GBM), and multinomial logistic regression with reported improvements in sensitivity, specificity, and accuracy.

Scientific Applications:

• Cancer molecular subtyping: Assigns molecular subtypes to cancer samples to characterize tumor heterogeneity.
• Colorectal cancer: Validated in case studies of colorectal cancer for subtype classification.
• Breast cancer: Validated in case studies of breast cancer for subtype classification.
• Precision oncology support: Provides single-sample subtype predictions that can inform treatment-related decisions in research and clinical contexts.

Methodology:

Supervised deep learning models are trained on gene expression profiles transformed into functional spectra that quantify biological pathway activities, and deep features are extracted for molecular subtype classification.