cola

cola implements consensus partitioning to classify samples into stable molecular subgroups by aggregating clustering results from multiple runs on random data subsets.

Key Features:

• General framework for consensus partitioning: Provides a configurable framework supporting feature selection, sample classification, and signature definition within consensus partitioning analyses.
• ATC feature extraction: Implements the Ability to Correlate to Other Rows (ATC) method for feature extraction, which was benchmarked to outperform other methods on public datasets.
• Subgroup classification with skmeans: Supports spherical k-means (skmeans) for subgroup classification, reported to yield better results than traditional clustering in benchmarks.
• Parameter benchmarking: Includes tools to benchmark key consensus partitioning parameters to guide parameter selection.
• Parallel application of multiple methods: Enables simultaneous application and direct comparison of multiple partitioning methods.
• Visualization capabilities: Provides visualization functions to assist interpretation of partitioning results.

Scientific Applications:

• Subgroup identification in genomic studies: Identifies stable subgroups within high-throughput genomic datasets to assess classification reliability.
• Cancer genomics: Facilitates discovery and validation of molecular subtypes in cancer datasets.
• Personalized medicine: Supports stratification of patients by molecular profiles for precision medicine applications.
• Evolutionary biology: Aids comparative analyses where subgroup structure in genomic data informs evolutionary questions.

Methodology:

Performs consensus partitioning by summarizing classifications from multiple clustering executions on random data subsets; implements ATC (Ability to Correlate to Other Rows) for feature extraction; uses spherical k-means (skmeans) for subgroup classification; provides parameter benchmarking and parallel application and comparison of multiple partitioning methods, with benchmarking on public datasets.