SMRT

SMRT performs integrative multi-omics cancer subtyping via a randomized transformation to identify molecular subtypes from profiles such as mRNA expression and methylation that associate with differential patient survival.

Key Features:

• Randomized Transformation (SMRT): Applies a randomized transformation to integrate multi-omics features prior to clustering.
• Scalable Analysis Pipeline: Integrates and analyzes large multi-omics datasets and can process hundreds of thousands of samples in minutes.
• Integration of Diverse Data Types: Handles data types with varying numbers of patients, including un-matched datasets, enabling joint analysis across disparate scales.
• Ensemble-based Perturbation Clustering: Uses ensemble perturbation clustering with memory-efficient implementations to derive robust subtypes under computational constraints.
• Benchmark Performance: Demonstrated superior performance relative to eight state-of-the-art methods in analyses of 37 TCGA and two METABRIC datasets (~12,000 samples across 28 cancer types) for identifying subtypes with significantly different survival profiles.

Scientific Applications:

• Cancer subtyping: Delineates molecular subtypes across cancers using integrated mRNA expression and methylation profiles.
• Survival-stratified subtype identification: Identifies subtypes that associate with significantly different patient survival profiles.
• Large-scale cohort analysis: Enables subtyping across large consortia datasets such as TCGA and METABRIC encompassing thousands of samples and multiple cancer types.

Methodology:

Applies a randomized transformation for feature integration followed by ensemble-based perturbation clustering, with scalable integration of multi-omics data and memory-efficient clustering implementations supporting unmatched sample counts.