segmentation fault after core.kinematics.FoldTree in homology modeling

Segmentation faults unfortunately aren’t error messages, so they don’t help us diagnose the problem (it means Rosetta crashed by wandering outside its allowed memory allocation). Can you post a command line?

Off the top of my head, the FoldTree is composed of a ton of tiny fragments. I don’t know if that’s appropriate for where you are in the protocol (but I’m guessing not, if it’s vanilla homology modeling – Rocco, can you correct me?)

320 residues is also awfully large for a homology modeling problem, which may be part of the problem.

-select_best_loop_from 1

If I understand this flag correctly, it does nothing when the argument is 1. An argument larger than one causes it to do extra modeling (N models) and then select a best sub-model from the N…but 1 model is 1 model.

-random_grow_loops_by 4

How close are your closest loops to each other in primary sequence? Is it possible that the loops could come into overlap because of the random growth (are the closest loops 8 or fewer residues apart in primary sequence?) Try it without this flag?

Posting your actual loops file may help us.

Foldtree manipulation does happen in homology modeling, particularly in the loop remodeling stages, where the loops are represented as branches in the FoldTree, relative to the fixed core. (This is to avoid leaver arm effects with moving the loops.) The FoldTree mostly looks like typical loop remodeling one (not surprising , given the protocols.looprelax lines), with polymer (-1) edges connecting the “fixed” residues, with jumps (positive numbers) from fixed region to fixed region across the loops, and then the loops being constructed from either side to the middle.

However, you get rather strange in the middle, where you have jump 4 across the loop from 123 to 208, then you have the cut loop from 123 to 190 and 208 to 191, but then you have a straight polymer run going from 208 to 123, and a third later on from 123 to 209, and then from 217 to 123 etc. It looks like the protocol is trying to model loops on loops and loops overlapping loops. My guess is that’s why it’s crashing. Are you specifying a loops file?

You haven’t really provided enough information to say. What is the full command line you used to start the run (including which application you used), and your full flag files? It’s best to just copy/paste the whole thing as-is, rather than trying to edit things, as you avoid things like “-loops:aat000_09_05.200_v1_3.gz” which would cause an early halt to the run because of bad options.

That said, the subset of flags shown probably isn’t the cause of the issue. As mentioned before, the core of the problem is probably how you’re specifying the loops or how Rosetta is auto-determining the loops.

I can’t see anything untoward with those flags offhand.

As long as your alignment file is set up correctly (and there is no additional issues), Rosetta should be able to handle unaligned regions. And by “handle” I mean “not crash on”, not necessarily “model accurately”. If you have large loops (e.g. larger than 10 aa or so) Rosetta has a problem, as it doesn’t have enough template information to model them correctly. Two large loop, especially ones interacting with each other, will likely get you a large number of possibilities, most of which are garbage.

If these domains are likely to be unimportant structurally with respect to the structure of the remaining protein (a reasonable possibility, given that the homologs don’t have them), then one possibility is to attempt to model the protein without those loops, by truncating them/replacing them with a shorter dummy loop. If the loops are thought to be an independently folding domain, you could try to model them by ab initio (each separately, as 100 aa is in the range to do so, but 200 is probably too big), or by homology modeling with homologs, if you have them – either separately or with dummy connecting loops if they’re a pair.

If there are homologs for the loops, albeit in different PDBs than the main portion of the protein, the new “hybrid protocol” might be useful to you. I believe (but am not certain) that it will be included in the upcoming 3.5 release. It will allow you to mix and match homologs for different sections of the protein to fold.

If you’re still getting segmentation faults, I’d probably recommend running in debug mode to see if there’s a more helpful error message. Past that, a debugger would be needed. (See https://www.rosettacommons.org/node/3226#comment-5904 and https://www.rosettacommons.org/node/3226#comment-5913 for specifics.)