MutSpace

MutSpace embeds large-scale sequence contexts to analyze cancer mutational signatures by capturing megabase-scale mutation rates and local mutational patterns to characterize somatic mutations and distinguish cancer subtypes.

Key Features:

• Large-Scale Context Embedding: Embeds broad sequence contexts to capture global mutation rates at megabase scales for mutational signature analysis.
• Patient-Specific Feature Extraction: Generates patient-specific mutational feature embeddings that capture both coding and non-coding somatic mutations.
• Integration of Global and Local Patterns: Combines megabase-scale mutation rates with local mutational patterns to represent both broad genomic structure and local sequence context.
• Subtype Identification Accuracy: Demonstrated high accuracy in identifying molecular subtypes in an analysis of 560 breast cancer patient samples.
• Superior Performance in Simulations: Outperformed previous methods in simulation evaluations characterizing mutational features from known patient subgroups.

Scientific Applications:

• Cancer Heterogeneity Analysis: Characterizes inter-patient variability in somatic mutation patterns to study cancer heterogeneity.
• Molecular Subtype Classification: Distinguishes molecular subtypes of cancer using embedded mutational features.
• Coding and Non-coding Mutation Integration: Integrates signals from coding and non-coding mutations to inform subtype definitions.
• Personalized Oncology Research: Provides patient-specific mutational representations applicable to personalized medicine and subtype-driven studies.

Methodology:

Implements an embedding framework that integrates megabase-scale mutation rates and local mutational patterns into patient-specific representations; performance was evaluated by simulation and application to 560 breast cancer patient samples.