|
What is SpectralNET?
SpectralNET is a web-based application that calculates graph theoretic metrics of input networks. Input networks can be of chemical-assay interactions, of protein-protein interactions,
or any other biochemically-based graph. In fact, even though this tool was originally intended to be
used solely with biochemical networks, it may be used to analyze the topology of just about any type of undirected graph.
The main tools of analysis that this application provides are:
- A global view of the graph drawn using the Fruchterman-Reingold algorithm.
- Detailed information about each vertex, including its degree, cluster coefficient, min/max/average distance, and corresponding values in the first few eigenvectors.
- Graphs of the degree distribution, degree vs. cluster coefficient, and degree vs. average path length for every graph.
- For a selected connected component with sufficiently many vertices, a graph of its eigenvalues, first few eigenvectors, inverse participation ratio, and spectral density. Also for sufficiently large connected components, the main graph can be redrawn using Laplacian eigenvectors one and two as axes. This is a form of local linear embedding using the graph Laplacian, which is described in Laplacian Eigenmaps for Dimensionality Reduction and Data Representation, Belkin and Niyogi, 2002.
- The ability to click graphs to see the vertex corresponding to the data point clicked, where this makes sense.
Please note that this application is intended to be used as a tool for initial investigation. We recommend that significant results be verified using other means prior to publication, since bugs may be present that we are unaware of.
This application was tested with Internet Explorer version 6. It is intended to be used on a screen with horizontal resolution of at least 1024 pixels.
Questions and comments should be sent to the author, Joshua Forman, at jforman@post.harvard.edu.
|
|