Potential new summary statistics #440
Description
There are some potential new summary statistics that could be useful when figuring out plausible vs implausible networks/parameters when we are doing the grid search. A lot of the statistics have to do with hub nodes, but it is undecided what is the best way to deal with hub nodes because it seems to be context specific. Maybe the solution is that we just should add a lot of different ways to deal with hub nodes.
Some come from the Automating parameter selection to avoid implausible biological pathway models paper under the Evaluating reconstructed pathway plausibility section.
- Hub node dependence
- Clustering coefficient
- Assortativity
Tristan also mentioned another statistic called the rich club coefficient that could be useful as well
Coefficient that's high when hub nodes in a network are also connected to each other. I remember reading in papers adjacent to BowTieBuilder that hub nodes in PPIs are usually not connected, so PPIs have a low rich club coefficient.