HCsnip

HCsnip implements guided piecewise snipping to decompose hierarchical clustering (HC) trees into non-overlapping clusters that integrate molecular data and clinical information for biologically and clinically relevant subgroup discovery.

Key Features:

• Semi-supervised piecewise snipping: Cuts the HC tree at variable heights using a semi-supervised framework to extract clusters beyond fixed-height methods.
• Integration of clinical information: Incorporates clinical data such as patient follow-up and survival data into the clustering objective to align clusters with clinical outcomes.
• Flexible data support: Supports a variety of data types for constructing and decomposing HC trees and the inclusion of auxiliary information like survival data.
• No predefined cluster number: Identifies cluster boundaries without requiring a user-specified number of clusters.
• Cluster quality evaluation: Provides functions to evaluate cluster quality using multiple criteria to assess consistency with molecular and clinical data.
• Significance testing and visualization: Implements permutation-based significance testing and visualization methods using sample molecular entropy.

Scientific Applications:

• Genomics subgroup discovery: Detects molecularly coherent subgroups within genomic datasets by decomposing HC trees.
• Clinical cohort stratification: Identifies clinically relevant patient subgroups by integrating patient follow-up and survival information into clustering.
• Translational and personalized medicine: Aligns molecular cluster structure with clinical outcomes to support interpretation relevant to personalized medicine.

Methodology:

Guided piecewise snipping performs variable-height cuts on hierarchical clustering (HC) trees in a semi-supervised manner integrating molecular and clinical data, with cluster quality evaluated by multiple criteria, significance assessed via permutation tests, and visualization using sample molecular entropy.