[NetBio] Precdicting network with modularity (which can be safely combined) #9
Description
Why predicting small networks can be hard
Well it's based on the nature of the algorithms. To infer connections, one typically aggregating data for many genes under many scenarios.. It's less often to make prediction for a smaller set of genes and combine them afterwards, since it's less likely to select the candidates reliably a priori. Because the interaction network is so sparse, if one just random chooses a subset of genes, it's possible that there are no edges between the chosen genes at all.
Hence it's impractical to build small network in the sense that you would always need a large dataset to generate any network prediction, so why don't just use the data fully. On the other hand, I am actually a big supporter of "network growing", which is similar to what you are trying to do.. Namely constructing a minimal graphical network that fully explains the phenomena. To this end, Chow-Liu tree is an excellent pioneer in using tree/hierarchical network to approximate the real interaction net.