I am trying to design a helical linker between two strongly interacting subunits that connects their termini using RosettaRemodel.
I am running remodel with the following flags in mpi mode and get the desired output (around 2-4 structures with -num_trajectory 32). However when I increase -num_trajectory to e.g. 1000 I still get only around 2-4 strucures out (named 1.pdb 2.pdb ...). while computation time increases drastically.
Is this low output expected in this case? Why am I not getting more output structures when more computation time / trajectories are used? Any information or ideas on how to increase the output for the larger calculations is welcome!
-in:file:s input/my_protein.pdb -remodel:blueprint input/resfile -run:chain A -remodel:num_trajectory 32 -use_clusters false -nstruct 1 -ex1 -ex2 -packing:extrachi_cutoff 5 -use_input_sc -packing:linmem_ig 10 -no_optH false -find_neighbors -hb_srbb 1.0
The file you care about has `_0001.pdb` suffix.
`-nstruct` controls how many times the program is run, giving independent structures. `-num_trajectory` controls how many trials the MCMC does to find the best, which is spat out.
Thanks a lot for your reply.
The Remodel documentations states "CURRENTLY REMODEL HANDLES ITS OWN FILE I/O, and only use job_distributor to launch the process. Normally job distributor will write out a file at the end of a run, usually in the format of XXX_0001.pdb; this file is NOT TO BE TRUSTED IF YOU RUN -num_trajectory GREATER THAN 1!! Instead, look for the files that are simply 1.pdb, 2.pdb, etc. [...]" which is why I was looking for these files. Is this outdated?
I also tried to set -num_trajectory to 1 and higher -nstruct (e.g. 100) however all structures with `_0001.pdb` suffix just featured perfectly linear stretches protruding from the protein with no folding at all. So should I set -num_trajectory to a number where I start to see some meaningful output and then increase the number of calculations using -nstruct?