hubViz

hubViz visualizes hub-centric structure in binary genomic data using a latent space joint model (LSJM) to position hubs centrally and improve interpretation of hubness in biological datasets.

Key Features:

• Latent Space Joint Model (LSJM): A model explicitly tailored for hub-centric visualization that positions hubs at the center of the latent space.
• Binary Genomic Data Support: Designed to operate on binary datasets and to emphasize hub structures within such data.
• Contrast with MDS / t-SNE / UMAP: Addresses limitations of Multidimensional Scaling (MDS), t-Distributed Stochastic Neighbor Embedding (t-SNE), and Uniform Manifold Approximation and Projection (UMAP) when applied to binary data and hubness.
• Hub-Centric Layout: Arranges central hub observations centrally with contrasting elements distributed around them to facilitate visual interpretation of connectivity.
• Simulation Validation: Evaluated through simulation studies that confirmed its effectiveness for clear visual inspection of hub structures in binary data.
• Implementation: Implemented as an R package and described as computationally efficient.

Scientific Applications:

• Gene Expression Data: Applied to gene expression profiles from multiple brain regions in rats exposed to cocaine.
• Single-Cell RNA-seq Data: Applied to peripheral blood mononuclear cells (PBMCs) treated with interferon beta.
• Literature Mining / Disease Networks: Used to investigate relationships among diseases derived from literature-mining datasets.

Methodology:

Computational methods explicitly include a latent space joint model (LSJM) for hub-centric latent-space positioning, evaluation via simulation studies, and implementation as an R package.