cancerAlign

cancerAlign aligns somatic mutation profiles across multiple cancer types using adversarial learning to enable tumor stratification and cancer gene prioritization.

Key Features:

• Cross-Cancer Type Alignment: Maps tumors from a target cancer type into latent spaces derived from other source cancer types to reveal shared patterns across cancers.
• Adversarial Learning: Uses population-level adversarial learning to perform alignment without requiring paired or aligned tumor samples across cancer types.
• Consensus Clustering: Integrates cluster labels from multiple source cancer-type latent spaces via consensus clustering to derive robust tumor subtypes.
• Unsupervised Approach: Operates without predefined labels or annotations and is applicable to genome-scale molecular data such as somatic mutation profiles.

Scientific Applications:

• Tumor Stratification: Identifies biologically meaningful tumor subtypes by clustering aligned somatic mutation profiles across cancer types.
• Cancer Gene Prioritization: Improves prioritization of cancer-associated genes by leveraging transferable mutation patterns across cancers.
• Transferability Insights: Assesses similarity and transferability of tumor characteristics across cancer types based on somatic mutation profiles.

Methodology:

Implements population-level adversarial learning to map target tumors into latent spaces of source cancer types, performs clustering within those latent spaces, and integrates cluster labels through consensus clustering; implemented in PyTorch.