RDRToolbox

RDRToolbox performs nonlinear dimensionality reduction and related analyses to reveal structure in high-dimensional genomics and microarray gene expression data.

Key Features:

• Nonlinear Dimension Reduction Algorithms: Implements Isomap (Isometric Mapping) and Locally Linear Embedding (LLE) to reduce dimensionality while preserving global geodesic structure and local relationships, respectively.
• Cluster Validation: Computes the Davis-Bouldin Index for assessing intra-cluster similarity and inter-cluster separation.
• Data Visualization: Provides plotting routines to visualize reduced-dimension embeddings.
• Data Generation: Includes generators for microarray gene expression datasets and the Swiss Roll dataset for benchmarking dimension-reduction methods.

Scientific Applications:

• Gene expression analysis: Analyze high-dimensional microarray gene expression profiles and other genomics data using nonlinear embeddings.
• Biomarker identification: Reveal patterns in reduced-dimensional space that can support identification of candidate biomarkers.
• Genetic network exploration: Uncover structure relevant to understanding genetic networks through preserved local and global relationships.
• Method benchmarking: Use the Swiss Roll and generated microarray datasets as benchmarks for evaluating dimension-reduction methods.

Methodology:

Isomap constructs a geodesic distance matrix from nearest neighbors and embeds data using that matrix; LLE reconstructs each data point as a linear combination of its neighbors to preserve local structure; the Davis-Bouldin Index is computed for cluster validation; data generators produce microarray gene expression and Swiss Roll datasets and plotting routines visualize embeddings.