Constructing Networks from Matrices
Methods for defining an adjacency matrix are also known as network construction-, inference-, or deconvolution methods. The network adjacency matrix can be defined by transforming a similarity or dissimilarity matrix, a symmetric matrix, or even a general square matrix. Multiple similarity matrices can be combined into a single “consensus” network, which allows one to define consensus modules. A signed correlation network turns out to be rank-equivalent to a Euclidean-distance-based network between scaled vectors. Sample networks, which are often defined as distance-based networks, are useful for identifying outlying samples or observations. A distance-based adjacency function yields an adjacency matrix A for which \(dissA = 1 - A\) satisfies the triangle inequality and other distance properties. Distance-based adjacency functions are useful for generalizing the ARACNE algorithm to general networks. We describe how the Kullback–Leibler pre-dissimilarity can be used (a) for measuring the difference between matrices and (b) for network construction.
KeywordsAdjacency Matrix Symmetric Matrix Positive Definite Matrix Dissimilarity Measure Dissimilarity Matrix
- Oldham MC, Langfelder P, Horvath S (2011) Sample networks for enhancing cluster analysis of genomic data: Application to huntington’s disease. Technical ReportGoogle Scholar