On 2/2/10 8:25 AM, abdullah ahmed wrote: > My question is related to optimization. I increased the default > optimization in "model-full.py" using the following code: > > # High VTFM optimization: > a.library_schedule = autosched.slow > a.max_var_iterations = 5000 > > Optimization does occur, however a few amino acids are changed from > their L-isomeric forms to the D-isomers.
Modeller contains a full description of the topology of the standard amino acids; the internal coordinates of these amino acids are for the L-isomers, so it will not produce D forms by default. However, since every optimization starts with a randomization, it is possible that that randomization produced a D-like residue which the optimizer is then unable to recover from, since it's a local minimum. This can happen at any optimization level.
The solution for problems like these is generally to build multiple models and then detect and discard "bad" models using the Modeller scoring function or an assessment score.
Ben Webb, Modeller Caretaker