Reproducing the exact result using information from previos run

Member Site Forums Rosetta 3 Rosetta 3 – Applications Reproducing the exact result using information from previos run

Viewing 1 reply thread
  • Author
    • #1530

        Hi Guys,
        I am currently trying to run few jobs using different set of parameters and I want to study the effect of those parameters on the results. How can I run a job using information of the previous run’s log file so that effect are only due to new parameters, not due to random simulation, as the no of decoys generated is very small and I feel though not very critical for the process, but for comparison purpose it’s important to the trajectories in the exact way as done in previous run.
        Please help me to solve this issue.


      • #8487

          You can fix your RNG seed with the flags:

          -jran #######

          Where ### is your seed number. (-constant_seed alone gets you a seed of 1111111). Your seed should be reported at the top of each log file. Fixing the RNG seed ensures the random number generator takes an identical trajectory.

          Notice that changing parameters for the simulation will necessarily drive you down a different Monte Carlo path – the RNG trajectory will be unchanged, but Monte Carlo will make different accept/reject choices because you will have different scores due to the new parameters. There is no way to ask Rosetta to, say, “always make exactly the same rotamer substitutions” if you have changed other parts of scoring, etc., so it might not be possible to do just what you want. Which parameters are you changing?

        • #8488

            Hi Steven!
            Thanks for your reply. I understand what you said. I just included few constraints and also used score12prime. But as you said all is random here, so what I wanted is not possible in MC simulation. Reaching to convergence will be the key to make a reasonable comparison between between different parameters.


        Viewing 1 reply thread
        • You must be logged in to reply to this topic.