wMKL

wMKL performs weighted multi-kernel learning to integrate heterogeneous multi-omics datasets and identify cancer subtypes by learning kernel weights and combining weight-related P-values.

Key Features:

• Integration of Heterogeneous Data Types: Models and integrates diverse multi-omics datasets to capture molecular heterogeneity in cancer.
• Flexible Weight Functions: Implements weight functions that incorporate prior knowledge into kernel weighting.
• Omnibus Combination Strategy: Combines weight-related P-values across weight functions using an omnibus strategy to synthesize contributions.
• Multiple Kernel Choices: Represents each data type with multiple kernel options to mitigate sensitivity to kernel parameter selection.
• Learning Weights of Different Kernels: Learns weights for kernels derived from each data type to reflect their heterogeneous contributions to subtyping.

Scientific Applications:

• Performance Benchmarking: Demonstrated superior performance relative to existing weighted and non-weighted methods in simulations and benchmarks.
• TCGA Applications: Applied to two TCGA datasets to identify novel cancer subtypes.
• Downstream Bioinformatics Analyses: Enables downstream analyses to investigate molecular mechanisms distinguishing the identified subtypes.

Methodology:

Each data type is modeled with multiple kernel choices; kernel weights are learned per data type and integrated across kernels; an omnibus combination strategy aggregates weight-related P-values to refine subtype identification.