CNSistent

CNSistent integrates and analyzes Somatic Copy Number Alterations (SCNAs) across heterogeneous cancer cohorts to discover recurring SCNA patterns and enable classification and explanatory analysis of cancer genomes.

Key Features:

• Imputation and Filtering: Provides imputation techniques to handle missing data within SCNA profiles.
• Consistent Segmentation: Employs consistent segmentation algorithms to delineate regions of DNA gain or loss across samples.
• Feature Extraction: Extracts features from segmented copy number profiles for downstream analyses.
• Visualization: Produces visualizations for interpretation of complex SCNA data.
• Integration Across Cohorts: Integrates SCNA profiles from cohorts such as TCGA, PCAWG, and TRACERx to increase dataset size.
• Machine Learning Applications: Supports deep convolutional neural network-based classification, demonstrates improved performance when trained on integrated datasets, and facilitates transfer learning between cohorts.
• Explanatory Analysis: Applies integrated gradients to attribute contributions of specific SCNAs to classification outcomes, for example highlighting SOX2 amplifications in lung squamous cell carcinoma.

Scientific Applications:

• Recurring SCNA discovery: Identification of SCNA patterns that arise from shared selectional pressures or common mutational processes.
• Cross-cohort genomic studies: Combination of TCGA, PCAWG, and TRACERx data to enable large-scale analyses of copy number alterations.
• Cancer classification: Classification of cancer types and subtypes using deep convolutional neural networks trained on SCNA features.
• Interpretation of drivers: Attribution of driver SCNAs, such as SOX2 amplification in lung squamous cell carcinoma, using integrated gradients.
• Transfer learning: Transfer of models between cohorts to leverage integrated-dataset training for other datasets.
• Oncogenic mechanism investigation: Analysis of SCNAs relevant to proliferation, tumor suppressor loss, and cellular immortalization.

Methodology:

Imputation of missing SCNA data; consistent segmentation of copy number profiles; feature extraction from segmented profiles; integration of SCNA profiles across cohorts (TCGA, PCAWG, TRACERx); deep convolutional neural network-based classification and transfer learning; explanatory analysis using integrated gradients; visualization of SCNA data.