The cutoff is used for clustering; it has nothing to do with an average structure (which itself is rarely, if ever, useful). The cutoff for clustering also depends strongly on what the system is and what you're considering for clustering. There is no generic or standard value.

I have a 50 ns simulation of protein-ligand complex. So, I would like to select a a representative structure to describe intermolecular interactions. What is the correct way to do it?

If possible, I have another question. What are the appropriate metrics that ensure that my system has equilibrated? 

Best regards.

Unfortunately "protein-ligand complex" is very vague and can range in meaning significantly. The cutoff for clustering is determined by a bit of trial and error. It depends on the size of the protein, its structural contents (if it's got a lot of loops you're going to have a lot of noise), etc.

There are plenty of articles and resources about convergence. What you need to demonstrate is that your results (whatever is useful and pertinent to the scientific question at hand) are invariant with time (within statistical error).

Dear Justin Lemkul, 

First of all, thanks for your prompt reply.

My system is a globular protein with 288 residues (30Kda) and a micromolecule (250Da). I have used gromos method with 0.2 nm. I have tried this criterion: best cutoff (10 < number of clusters < 100). What do you think about it?

Again, I am very grateful for your help.

Best regards.

For avg structure u can use chimera. and for selected structure for interaction use VMD