You get different values because packstat has a large random component.
You can use -packstat::oversample 100 to cause it to "run more" and be less random (I think this makes it run 100 times more, up from 1 cycle). packstat is slow, although since this is a run-once step at the end of a run it won't matter much. That will reduce the variance significantly.
Packstat is not well documented. It's searching for the presence of non-solvent-accessible "voids". It reports on the range 0-1, where 1 is perfect. I've seen "above 0.65" marked as "good enough". The scale was set with high-res crystal structures.
I usually turn packstat off for InterfaceAnalyzer - a lot of the things IA reports are there to test if they are useful; I found that packstat was not.
See also Protein Sci. 2009 Jan;18(1):229-39, which is the RosettaHoles paper. Packstat is "RosettaHoles lite" (especially light in that it does not require an external secondary executable, dalphaball.
You get different values because packstat has a large random component.
You can use -packstat::oversample 100 to cause it to "run more" and be less random (I think this makes it run 100 times more, up from 1 cycle). packstat is slow, although since this is a run-once step at the end of a run it won't matter much. That will reduce the variance significantly.
Packstat is not well documented. It's searching for the presence of non-solvent-accessible "voids". It reports on the range 0-1, where 1 is perfect. I've seen "above 0.65" marked as "good enough". The scale was set with high-res crystal structures.
I usually turn packstat off for InterfaceAnalyzer - a lot of the things IA reports are there to test if they are useful; I found that packstat was not.
See also Protein Sci. 2009 Jan;18(1):229-39, which is the RosettaHoles paper. Packstat is "RosettaHoles lite" (especially light in that it does not require an external secondary executable, dalphaball.