Thanks! With the multi-refined models, how can I make an average map for all the models? Can I do this in phenix? Fengyun 引用 Pavel Afonine ：

Hi Fengyun,

- to run many SA with different random seeds:

for random_seed in random_seeds: phenix.refine model.pdb data.mtz simulated_annealing=true main.random_seed=random_seed output.prefix=random_seed

To create even broader variety, I would slightly shake the starting model:

for random_seed in random_seeds: phenix.refine model.pdb data.mtz simulated_annealing=true main.random_seed=random_seed output.prefix=random_seed modify_start_model.sites.shake=0.3

Note, you will need to filter the outliers - refined models that have significantly higher R-factors compared to the rest.

...

I read on the paper that they use CNS (model_map.inp) to calculate the avarage sigmaA-weighted 2fofc map. Right now I have set up the simulated annealing with different random seeds. I wonder whether phenix could calculate the average map with the multiple coordinates from different starts?

Thanks Pavel! I am now running your script. 引用 Pavel Afonine ：

Hi Fengyun,

my previous email actually doens't answer your question - I realized this after I pushed Sent button -:)

To really answer your question I just wrote a Python script that does the following: - run multi-start Simulated Annealing, - combine all refined models into one multi-model PDB file (models separated by MODEL-ENDMDL records), and - compute averaged 2mFo-DFc map.

The complete working example is here:

http://cci.lbl.gov/~afonine/mssa/

You have two options at this point:

- take this example, slightly change inputs by editing the run.py file (change input data and model file names, number of SA runs, etc) and run it as following:

phenix.python run.py

(since I spent 15 minutes on writing this script it's obviously not thoroughly tested or parameter-optimized, although I believe it should do the job right)

or

- wait a few days for one of the next PHENIX nightly builds where the above script will be available as a user-friendly either phenix.multi_start_sa command or an option of phenix.maps (I haven't decided yet what is better).

Let me know if you have any questions.

Pavel.

...

I read on the paper that they use CNS (model_map.inp) to calculate the avarage sigmaA-weighted 2fofc map. Right now I have set up the simulated annealing with different random seeds. I wonder whether phenix could calculate the average map with the multiple coordinates from different starts?