I have a question about how atom coordinates are determined for predicted structures during the ab inito folding protocol and with the extraction of the PDB of the predicted structure from the output silent file.
I am currently working on development of a new score term, called ms_labeling, to be added to Rosetta. This term is based off of experimental mass spec covalent labeling data. In short, the score is calculated by counting the number of neighboring atoms within a certain radius of each LYS. This atom count is compared to an input file containing experimental data and each LYS is scored either 0 or -1 based upon its agreement with the input data (this is a really rough model). I have a working code that correctly counts the atom number of neighbors and I have been able to implement this into Rosetta and run ab initio folding and calculate total Rosetta scores including my new score term. The problem I am running into is that for a few of the predicted structures, if I extract the PDB from the silent file and rescore the structure using either the ./score.linuxreleasegcc application or a simple application that I wrote, the ms_labeling score term that I added gives different results. I don't understand how this could be possible since the since it is based strictly on the atom count.
I'm wondering whether there is some sort of subtle conformationaly change that occurs between the structure that gets scored during the ab initio calculation and the structure that can be extracted from the output silent file. I've tried to find an answer to this, but haven't had much luck. Any insight would be very much appreciated!
Are your ab initio structures centroid mode or fullatom mode? Last I knew, ab initio output in centroid. If score is turning centroid to fullatom, that will upset the number of atoms greatly.
The other obvious alternative is repacking: if structures are getting repacked (especially if there are missing sidechains, but that's essentially impossible from Rosetta fullatom output), that will change the number of atom neighbors.
Finally, Rosetta scoring is not identical between the output of a structure and the rereading of the same structure...with PDBs. That's because they aren't precise to enough digits compared to the sensitivity of fa_rep and other terms. Depending on your type of silent file I guess it could be a re-read precision-related issue.
If you're in a Rosetta lab (you did say "to be added to Rosetta"), this should just go on the mailing list.
Thank you so much! I am a part of a Rosetta lab and I honestly didn't think about putting this on the mailing list (sorry, this whole thing is still pretty new to me). At this point I have just been working with my own copy of the source code. I haven't done any uploading with Git or anything.
I think that this might answer my question though. After looking into it a bit more, it does look like ab initio uses a centroid score function by default.
I'll repost this to the mailing list.
Thanks!