Re: [modeller_usage] automodel optimization

[Modeller Caretaker <modeller-care@ucsf.edu>]

Alex Brown wrote:
> I would presume that automodel uses model.optimize when optimizing a 
> model, with a default of a single pass, automodel.repeat_optimization 
> having a default value of 1.

Actually it does one pass through the schedule (see 
http://salilab.org/modeller/8v2/manual/node147.html) which in turn does 
several optimize calls. These are generally conjugate gradient 
optimizations.

> However, according to the manual, the 
> optimization_method parameter of model.optimize has default value of 
> 999, whereas it should have a value of either 1 or 3. What default 
> optimization method does automodel use, can it be changed, and what does 
> the model.optimize default of 999 mean ?

The default value of 999 instructs Modeller to read the optimization 
method from the previously-defined optimization schedule instead. You 
can tweak the schedule a little with automodel.max_var_iterations and 
automodel.repeat_optimization (see 
http://salilab.org/modeller/8v2/manual/node36.html) or you can replace 
it entirely by setting library_schedule (see 
http://salilab.org/modeller/8v2/manual/node37.html).

After the schedule-controlled CG optimization, the model is refined by 
MD simulated annealing, which can be controlled with the 
automodel.md_level variable. The procedure is outlined in the Modeller 
papers.

	Ben Webb, Modeller Caretaker
--
modeller-care@ucsf.edu             http://www.salilab.org/modeller/
Modeller mail list: http://salilab.org/mailman/listinfo/modeller_usage