KLIC

KLIC performs integrative consensus clustering of heterogeneous multi-omic datasets using multiple kernel learning to weight and combine data modalities for tasks such as tumor subtyping and transcriptional module discovery.

Key Features:

• Multiple Kernel Learning Framework: Employs multiple kernel learning to integrate diverse datasets by assigning dataset-specific weights based on informativeness and relevance.
• Weighted Contribution: Dynamically adjusts the contribution of each dataset so noisy or less informative datasets are down-weighted in the clustering outcome.
• Integrative Consensus Clustering: Produces a consensus clustering solution that combines information across high-dimensional omic datasets.
• Benchmarking Against COCA: Has been systematically benchmarked against Cluster Of Clusters Analysis (COCA), demonstrating superior performance in scenarios including noisy data.
• Simulation Studies: Performance has been evaluated through extensive simulation studies across different data conditions.
• Applications in Cancer Research: Tailored for cancer-related analyses such as tumor subtyping and transcriptional module discovery using omic data.

Scientific Applications:

• Tumor Subtyping: Integrates multiple omic data types (e.g., genomics, transcriptomics) to identify and refine clinically relevant tumor subtypes.
• Transcriptional Module Discovery: Discovers gene expression modules and transcriptional patterns by combining signals across datasets.

Methodology:

KLIC frames integrative clustering as a multiple kernel learning problem, allowing different datasets to contribute variably to the final clustering solution based on their quality and relevance so that more informative datasets have greater impact while minimizing the influence of noise.