multiview

multiview implements multiview extensions of dimensionality reduction and spectral clustering algorithms to integrate heterogeneous biological data and enable improved visualization and clustering of high-dimensional multi-view datasets.

Key Features:

• Multiview t-Distributed Stochastic Neighbor Embedding (t-SNE): Adapts t-SNE to handle multiple views, reducing high-dimensional data to two or three dimensions while preserving local structure across views.
• Multiview Multidimensional Scaling (MDS): Extends MDS to represent multiview data in a lower-dimensional space while maintaining pairwise distances across different views.
• Multiview Minimum Curvilinearity Embedding: Embeds multiview data into a low-dimensional space minimizing curvilinearity distortion to preserve intrinsic dataset geometry.
• Multiview Spectral Clustering Method: Applies spectral clustering tailored for multiview datasets, leveraging information from multiple sources to identify clusters and often yielding superior results versus single-view methods.

Scientific Applications:

• Integrative analysis of multi-omics data (genomics, proteomics): Enables joint analysis of genomics and proteomics views to integrate complementary molecular information.
• Clustering and visualization of complex biological datasets: Provides multiview-based clustering and low-dimensional embeddings to improve interpretation of heterogeneous biological samples and patterns.

Methodology:

The methods extend established single-view pattern recognition techniques to integrate information across multiple data views and were evaluated on four multiview datasets where they outperformed single-view counterparts.