web-rMKL

web-rMKL performs integrative dimensionality reduction and clustering of multi-omic, multi-view datasets using unsupervised multiple kernel learning (rMKL-LPP) to identify sample groups such as cancer subtypes and cell differentiation states.

Key Features:

• Dimensionality reduction and clustering: Integrates dimensionality reduction with subsequent clustering of samples derived from multiple input datasets to handle high-dimensional multi-omic data.
• Unsupervised multiple kernel learning (rMKL-LPP): Implements rMKL-LPP to capture complex relationships across different data views and has demonstrated superior performance in clinical enrichment benchmarks of multi-view clustering algorithms.
• Kernel matrix preprocessing: Generates kernel matrices from tables containing numerical feature values when precomputed kernel matrices are not available.
• Multi-view data support: Operates on heterogeneous molecular views including genetic data and histone modification profiles for integrative analysis.

Scientific Applications:

• Cancer subtype discovery: Integrates diverse molecular features to identify and characterize cancer subtypes.
• Stem cell differentiation studies: Applies multi-view integration to analyze molecular changes during stem cell differentiation.
• Clinical enrichment analyses: Supports analyses aimed at clinical enrichment and has shown improved performance in benchmark comparisons of multi-view clustering methods.

Methodology:

Integrates multiple data sources via unsupervised multiple kernel learning using rMKL-LPP to reduce dimensionality while preserving essential information and enable clustering, with optional preprocessing to produce kernel matrices from numerical feature tables.