# Rosetta 3.4 installation on fedora core 14

24 posts / 0 new
Rosetta 3.4 installation on fedora core 14
#1

Hi,

I am trying to install Rosetta3.4 on fedora 14 platform. I am getting the following error after issuing the command : scons bin mode=release

ObjexxFCL -lz -lcppdb -lsqlite3
/usr/bin/ld: cannot find -lz
collect2: ld returned 1 exit status
scons: *** [build/src/release/linux/2.6/64/x86/gcc/4.5/libutility.so] Error 1
scons: building terminated because of errors.

Pls guide ??

Post Situation:
Thu, 2012-03-29 16:50
bharat_46010
Thu, 2012-03-29 17:30
smlewis

Thanks for the help. While installation is going on I want to ask whether I can use pmutscan to search for the stable mutants. As in my case I developed a library of mutants , and I want to know what all mutant would be stable as per rosetta.

Thu, 2012-03-29 18:28
bharat_46010

Use pmutscan (which I'm told is released; I never checked) if you want single or double mutations that increase stability. Use fixbb if you want more aggressive (larger) numbers of mutations.

Thu, 2012-03-29 18:40
smlewis

pmut_scan is really meant to be used as a first-pass scan for stabilizing single point mutants and/or double mutants. If you already have a set of single or double mutants in mind, the ddG_monomer protocol is the more accurate way of estimating mutant ddGs using Rosetta. See the documentation for the ddG monomer application (http://www.rosettacommons.org/manuals/archive/rosetta3.4_user_guide/d3/d...) for more info.

Mon, 2012-04-02 07:13
ronj

Hi,

I want to know if I recompile Rosetta3.4 with mpi , can I continue with the calculation that has been done so far with Rosetta3.4 without the mpi. Since my system is having 24 processors and it seems that Rosetta is using only 2 of them . Also what extra option I need to include while running any protocol ?

Tue, 2012-04-03 15:48
bharat_46010

It depends on the application, and to some degree on the options you give it. I'm going to guess pmutscan will not allow you to do this, as it's not a random many-trajectories Monte Carlo search like most applications are, and doesn't run through a standard job distributor. Which application?

Tue, 2012-04-03 16:41
smlewis

I want to run loop_modelin protocol with kic option.

Tue, 2012-04-03 19:44
bharat_46010

I did a quick test and it looks like loop modeling is one of the executables that don't work right in this regard (it hasn't been switched to JD2 yet). Sorry.

If you just start the MPI job in a new directory, the results can be combined later; you'll just have to keep an extra decorator in their name (probably the directory name/number) to keep the different structure IDs separate.

Wed, 2012-04-04 10:07
smlewis

I have already created a separate file for each resfile and it's corresponding structure. Will that work ?? . For compilation I need to include extras=mpi , but while running the protocol , what extra option I need to include.

Wed, 2012-04-04 16:24
bharat_46010

You will need a loop file for each structure, and I would use Jran to be safe (see the abinitio modeling manual), but smlewis might know if this is needed or not. I always use it just in case. Make sure to use -ex1 -ex2 and -loops:max_kic_build_attempts 10000, as you will have missing output structures otherwise (even with small loops the default 100 is just not enough.) Don't forget to use the extended option in the loop file if you want to throw away the initial loop!

Since loop modeling is not yet JD2 compatible, you will need to make a directory for each structure that you are running loop modeling on. CD into that directory, and run the loop modeling application from there. This is for each processor that you plan on using. Since it is JD1, it is stupid, and will try to overwrite any files that you have in the directory. Later, combine all directories or use the linux find command to get all filenames for use in subsequent analysis/etc. I always loose the score information, so you may want to write something to save and combine that. Not sure if a script exists somewhere. I have some python scripts you could use if you are using QSUB...you would have to edit them, but I could help....

Wed, 2012-04-04 19:17

Thanks for the reply. In the present situation, I am doing the loop_modeling in the same way that you have just mentioned. I created a separate directory for each structure and it's resfile and then ran loop_modeling for each directory. I am generating 10 model for each structure , as the number of resfile are many in my case. This means that I can continue with the same but with additional options for mpi. I am having 24 processors in my system. In addition to that loop modeling has to be done for 220 structures with 400 resfiles. What options I need to use in such a case ??

Wed, 2012-04-04 19:55
bharat_46010

To be honest, I havn't used MPI. I was happy to just have Rosetta compiled and working on our cluster! The rest of the things needed to run should be MPI specific. Here is a script I use to run NAMD which should give you an idea...

#!/bin/bash
#PBS -q dna
#PBS -l nodes=10:ppn=16
#PBS -N MIS_NPT_Proc_Test
MPIEXEC=/apps/mpich2/bin/mpiexec
PPN=16
echo cat $PBS_NODEFILE awk '{if ($0 in vett) print $0":"2; vett[$0]="x"}' $PBS_NODEFILE >>/common/madsci/Dynamics/NAMD/Run/jobs/$PBS_JOBID.nodefile

NPROCS=wc -l < $PBS_NODEFILE echo This job has allocated$NPROCS nodes

NODES=$((NPROCS/PPN)) echo total nodes check$NODES
mpiexec -machinefile /common/madsci/Dynamics/NAMD/Run/jobs/$PBS_JOBID.nodefile -np$NPROCS /common/madsci/Dynamics/NAMD/Program/NAMD_2.7_Source/Linux-x86_64-g++/namd2 /common/madsci/Dynamics/NAMD/Configs/personal/MIS/MIS_NPT.conf> /common/madsci/Dynamics/NAMD/Configs/personal/MIS/MIS_NPT_10ns_test_short.out
#-machinefile /common/madsci/Dynamics/NAMD/Run/$PBS_JOBID.nodefile -np$NPROCS
#/apps/mpich2/bin/mpdboot --totalnum=$NODES --ncpus=$PPN --file=/common/madsci/Dynamics/NAMD/Run/\$PBS_JOBID.nodefile
#/apps/mpich2/bin/mpdallexit

Thu, 2012-04-05 08:02

While the resfile option will take more than one resfile, few applications ever use more than the first resfile offered. If you want to use a variety of resfiles, you can't offer them to Rosetta all at once, which effectively means you cannot use MPI anyway, and should just run separate, simultaneous runs in different directories.

Thu, 2012-04-05 09:13
smlewis

I think my situation is similar to what you just said and I am doing the same at present. Anyways thanks for your replies ..

Thu, 2012-04-05 21:39
bharat_46010

Loop modeling is MPI-enabled, but not through JD2. Compiling with MPI tells Rosetta all it needs to know. I don't know and can't help with your MPI setup otherwise - for me I just add "mpirun" before Rosetta as part of the command line, and it works, but that may vary for your MPI installation.

Thu, 2012-04-05 09:11
smlewis

Hi,

I tried with the calculation of ddG between the wild type and the mutant. Here's what I did :-

1. I relaxed the wild-type beta-hairpin structure , generated some 50 structures and chose the lowest scoring for further loop refinement. For loop refinement I generated some 50 structures and again chose the lowest scoring one. Then I calculated the Delta G value for only one loop refined structure with the lowest energy.
2. The same thing I did for my mutants and calculated their delta G. Finally I calculated the difference i.e. ddG (delta G of mutant - Delta G of wild type)
3. I didn't get even a single structure out of 400 different structures that were highly stable than the wild type. However, I expected of getting atleast one or two structures highly stable than wild-type.

Does that mean that my method is wrong. As I repeated the same method for an experimental data where one of the mutant was more stable than the wild-type. But, in that case also Rosetta gave an opposite result. How could this be justified ??

Mon, 2012-04-09 06:41
bharat_46010

Ok, so a few things might help. First, look into the sequence tolerance application, http://www.rosettacommons.org/manuals/archive/rosetta3.4_user_guide/d1/d... and check out the referenced papers. Might be more in line to solve your problem.

Second, 50 structures for loop modeling is way to small to produce anything worthwhile. If you don't have computational resources, at least try one or two hundred structures for each design. 1000 would be ideal. How long is your loop? Single loop or multiple? How many mutations are in the loop? (just one, and you may be able to get better results by backrub or relaxing the loop). Are you using KIC for low-res and high-res? (high-res kic does not change much at all) Are you keeping the original backbone of the loop - ie not 'extended' in the loop file?

Mon, 2012-04-09 07:54

Actually I have also tried with generating 1000 structures, but I found that the score did not deviate much. That's why I decided to generate 50 structures. The loop is 4 residue long. It's a single loop with only two mutations. I am using kic option for refining the loops. The original backbone of the loop is extended . Here's the loop file's contents : LOOP 11 14 13 0 1.

Mon, 2012-04-09 16:32
bharat_46010

Ok, were getting to the end of my knowledge, but I will try to help. Thanks for the info. First, why 4 residues? Try increasing the loop size. With the cutpoint a residue away, you may not be able to sample many structures. If there is SS anchoring the loop, then I would relax or backrub the structures, as that may be a better simulation of the effect of the mutations. You are trying to see if these mutations can increase the stability of the protein? Is the loop at the surface? Is it binding to anything? Why only these mutations and these sites? Is this your benchmark? Why not do fixedBB or sequence tolerance experiments on the sites of the loop?

Some other suggestions - You could try starting without the extended configuration. You could try using the 2010 rotamer library, option -dun10. You could also try using the energy term hack_elec instead of fa_pair (Open up score12_full.wts from database/core/scoring/weights, and save a new file with hack_elec at the same weight as fa_pair. It is a coulombic electrostatic term researchers use for ligands that we have had success with. You could also try using the neighbor dependent ramachandran maps for scoring, option -score:ramaneighbors.

Perhaps someone else would have better advice? I know that it would be difficult to test all those things. Best bet is to relax or backrub these loops. There is not much backbone space to be sampled, and I am not convinced high-res KIC samples the backbone space well.

Mon, 2012-04-09 21:09

Actually I generated the structures for all those loop mutants by doing fixed bb design . Then I loop refined each mutant. Regarding increasing the size of loop, since loops are connecting two adjacent beta-strands and increasing the loop size would result in loosing some of the beta-strand residues, I guess ??. This study is to look at the effect of loop residues on protein stability. The loop is a surface exposed loop. It doesn't have any binding activity. The reason for studying mutations has specific location is the sole purpose of this experiment. No, we are not doing any benchmark study. I have tried with fixed bb first and was not happy with the result, as they were correlating with the experimental ones. I guess, I have not tried with sequence tolerance protocol, I will try it and let you know.

Well, I will try modeling without extending the loop and also with the new rotamer library. I don't think in my case I need any other scoring term. I think, scoring on basis of ramachandran map is a good option. Well, if you have said this I want to one more thing ...

I also want to classify these loop mutants as type I , II and type I' & II' on the basis of phi psi angle of 2nd and 3rd residue of a 4 residue loop. Since fixed bb design cannot be used in that as the main chain atoms are fixed in that case. So, I did loop refinement and chose the lowest scoring one to classify them further. Now as per the already established rules , NG i.e. residue 2 and 3 are always found in type I' loop (phi (2) - 60, psi (2) - 30 & phi(3) -90 , psi (3) - 0 , with +/-30 deviation for phi psi) , but in none of the structures I am getting NG in those phi psi angle range. Now this is an another issue ??

Regards
---------
BHARAT

Tue, 2012-04-10 16:16
bharat_46010

First, I am not familiar with beta-hairpin design, so I can't help you with the classifications.
However, regarding the experiment in general: You chose designs that best fit the backbone of the loop. Hence, now that you try to sample conformations of the loop, it is not changing much....It's sorta the problem of fixedBB design, you want mutations that will stabilize a protein or interface, but unless you conformationally sample all possible designs, you end up designing mutations that would best fit the backbone you started with...

People have had some success designing on ensembles, and some people will try to convince you that this is flexible backbone design. Technically, they are wrong, however, this is sorta the closest to flexible backbone design that I have found. It may be worthwhile to try. The newest sequence tolerance paper uses a few different types of ensembles including MD, and had decent results using the backrub motion for sampling...

Tue, 2012-04-10 21:42

Hi,

I tried using the backrub movement for one of my mutant structures. I generated 50 structures and each was given for 1000 trials. I am getting different scores for each structure. I found that thereś a huge difference amongst the fixed bb design score, loop refinement and backrub score. I want to know how to select the lowest scoring structure from amongst those generated after backrub, since score is not going to any file. Also, what should the ideal value (or rather the minimum value) for n struct and ntrials as I am having a large number of mutants . I did not understand the pivot residues concept and I tried giving residue 22,25 as pivot and in resfile I mentioned 22-25 residue no for backrub trial. I got some error. So, can u please tell me how to use this pivot residues.

Lowest score after fixed bb design : -24
Lowest score after loop refinement : -29
Lowest score after backrub : -18 (there was large variation in scores)

Mon, 2012-04-16 22:09
bharat_46010

You should read the sequence tolerance papers for nstruct/ntrials; backrub is a little different, and the original papers should cover that. There should be a large difference in score, as the backbone (and in design and loop modeling -rotamers) are different.

Tue, 2012-04-17 07:28