regulators are also in mRNAs list which gives a cyclic graph

[haluk]

I run the tutorial with its data. In module 2 CD53, and CAPS1 are regulators. They are also in the mRNAs.

How is this possible? Regulators are parents of mRNAs. In this case, we got a cyclic graph.

[haluk]

I saw the same problem in other modules too. Then, I thought maybe TFs are also included in the list for visualization purposes. But then, I found module 3 doesn't have this problem. GNA11 is only in TF list.

[haluk]

I think it's because assignRegulatorsNoAcyclStoch method is invoked in regulators task.

[erbon7]

That's indeed the default behavior of the program. However you have to remember that this is a test dataset. In real life, a transcription factor part of a given module could also be assigned as candidate regulator, but we do not often see that in practice. There might be a biological reason to that. More often than not regulators, such as TFs, do have a relatively "flat" profile and are not selected in gene clusters.

[haluk]

I'm confused. You are calling output as a module network. Underlying graph should be a Bayesian Network which by definition is acyclic. However, program gives cyclic graph in this example.

Is your module network definition different than the one introduced by Segal et al.?

[tmichoel]

The output network is a consensus network over many assignments of genes and regulators to modules, and as such need not be acyclic. In practice, we have decoupled module and regulator assignment completely and do not enforce acyclicity even at the level of individual solutions, because it leads to considerable computational speed-up, improved performance using empirical tests, and minor theoretical error. See this Bioinformatics publication for more details.

The original code for learning a single module network matching exactly the definition of Segal et al. is still included in the lemon-tree package, in case the underlying acyclic Bayesian network structure is important for you.