VISDA

VISDA applies coarse-to-fine (divisive) hierarchical clustering and hierarchical mixture modeling to identify and visualize clusters and informative genes in high-dimensional genomic datasets.

Key Features:

• Progressive Hierarchical Clustering: Uses a coarse-to-fine (divisive) hierarchical clustering approach that progressively refines clusters to reveal global and local data structure.
• Hierarchical Mixture Modeling: Integrates hierarchical mixture modeling to support clustering and to delineate cluster boundaries and relationships.
• Informative Gene Selection: Incorporates both supervised and unsupervised methods for selecting informative genes to focus analysis on biologically relevant variables.
• Data Visualization / Projection Methods: Employs multiple projection methods tailored to different clustering tendencies and supports supervised and unsupervised visualization techniques.
• User/Prior Knowledge Guidance: Incorporates user or expert knowledge to initialize models within low-dimensional visualization spaces and guide modeling choices.
• Model Order Selection: Applies Bayesian theoretic criteria for model order selection combined with user justification via a hierarchy of low-dimensional visualization subspaces.

Scientific Applications:

• Gene clustering: Identifies biologically relevant co-expressed gene clusters in genomic datasets.
• Sample clustering: Detects sample-level groupings in high-dimensional genomic data.
• Phenotype clustering: Captures pathological relationships among different phenotypes at the molecular level.
• Muscular dystrophy and muscle regeneration studies: Applied to muscular dystrophy and muscle regeneration data to identify biologically significant gene clusters.
• Multi-category cancer molecular phenotyping: Captures molecular-level relationships among phenotypes in multi-category cancer datasets.

Methodology:

Uses multiple local visualization subspaces corresponding to nodes of the hierarchical structure for subspace data modeling; implements coarse-to-fine (divisive) hierarchical clustering and hierarchical mixture modeling; applies supervised and unsupervised informative gene selection and multiple projection methods; initializes models using user/prior knowledge in low-dimensional visualization spaces and uses Bayesian theoretic criteria for model order selection.