VizStruct

VizStruct maps high-dimensional gene expression and single nucleotide polymorphism (SNP) datasets into lower-dimensional representations to enable exploration and pattern detection.

Key Features:

• Dimensionality Reduction: Uses complex-valued harmonics of the discrete Fourier transform (DFT) and the first Fourier harmonic projection to reduce multi-dimensional SNP and gene-expression data to two dimensions.
• Three-dimensional Mapping: Extends the DFT-based approach with Kullback-Leibler divergence to project multi-dimensional datasets into three dimensions (3D VizStruct).
• Pattern Recognition and Classification: Supports visualization-driven classification using the oblique decision tree algorithm OC1, with accuracy assessed by holdout and cross-validation.

Scientific Applications:

• Gene Expression Profiling: Applied to DNA array gene expression data to reveal expression patterns across thousands of genes.
• SNP Dataset Analysis: Applied to SNP genotype datasets including human Chromosome 21, the LPL gene locus, and coral population genotypes to reveal genotype characteristics.

Methodology:

Computational methods include complex-valued DFT harmonics and first Fourier harmonic projection for 2D mapping, a combination of DFT and Kullback-Leibler divergence for 3D mapping, visualization-driven classification with OC1 evaluated by holdout and cross-validation, and software implementations in Java and Matlab.