...
We captured the TOM matrices from the Sage and WGCNA algorithms. The values are the same (except for some small differences which look like rounding/accuracy differences) as illustrated in this scatter plot: 
Note: To make the TOM values the same for both algorithms, the default value for the WGCNA parameter "tomType" is overridden, changing it from "signed" to "unsigned". The definition of the parameter is given here: http://www.genetics.ucla.edu/labs/horvath/CoexpressionNetwork/Rpackages/WGCNA/OldPackages/WGCNA_0.67-2.pdf
- TOMType: a character string specifying TOM type to be calculated. One of "unsigned", "signed". If "unsigned", the standard TOM will be used (more generally, TOM function will receive the adjacency as input). If "signed", TOM will keep track of the sign of the adjacency between neighbors. ... If TOMType = "unsigned", entries of the adjacency matrix are required to lie between 0 and 1; for TOMType = "signed" they can be between -1 and 1.
The TOM values are the inputs to the clustering algorithms. The dendrograms appear nearly identical above, therefore it seems the differences between the two algorithms are in the module determination from the dendrograms.