In this paper, we focus instead on the biological relevancy of the resulting Relevance Networks.
This simulation study shows that Correlation Circle plots, Relevance Networks and CIM are able to highlight the relevant variables amongst the noisy ones and pinpoint the pair-wise association structure between the two data sets.
During his research under Isaac Kohane (at Children's Hospital) he developed a novel methodology for analyzing large data sets of RNA expression, called Relevance Networks.
Various refinements have been proposed to try to discriminate between direct and indirect interactions in relevance networks.
But in terms of identifying regulatory interactions, relevance networks are bound to be misleading because marginal independence alone cannot discriminate among direct and indirect dependencies.
However, in this challenge relevance networks have shown a better performance, both for the full data and for half the data.
This relation between the covariance inverse and the adjacency matrix links GMRF theory with graphical models, so extending the graphical models provided by relevance networks [8].
In particular, mutual information as used in Relevance Networks [10] or CLR [11] is a common dependency measure exploited in filter-kind approaches for feature selection [28].
Like Relevance Networks and CLR, this method reduces all the information contained in the expression data to mutual information between all pairs of genes, while our approach is by nature multivariate.
The overall scores for the relevance networks (31.64 for n = 100, and 26.30 for n = 50) are higher than those obtained with the graphical lasso (Table 2 and Table 3).
