diceR

diceR performs ensemble clustering to integrate results from multiple clustering algorithms for robust partitioning of high-throughput data and identification of patient sub-populations in cancer research.

Key Features:

• Ensemble Clustering Framework: Employs an ensemble clustering approach to integrate results from diverse clustering algorithms and mitigate individual algorithm biases.
• Algorithm Pooling Methods: Implements pooling methods including majority voting, K-Modes, LinkCluE, and CSPA (Cluster-based Similarity Partitioning Algorithm).
• Reduced Subjectivity in Model Selection: Provides procedures to guide algorithm selection and determination of the number of clusters to reduce subjective decision-making.
• Data-Agnostic Application: Applies to arbitrary datasets beyond biological contexts, enabling use across different data types.

Scientific Applications:

• Cancer Patient Stratification: Segments patients into sub-populations from high-throughput data to support diagnostic, prognostic, and therapeutic-response analyses in cancer research.
• Improved Generalizability and Reproducibility: Enhances generalization and reproducibility of clustering results across cohorts through ensemble-based consensus partitions.

Methodology:

The package performs ensemble clustering by pooling outputs from multiple algorithms using majority voting, K-Modes, LinkCluE, and CSPA, includes procedures for algorithm selection and choosing the number of clusters, and its tools are data-agnostic.