pritesh bhat wrote: > I am having some confusions regarding the number of model to be > generated to get good model from MODELLER. Exactly how many models > should be generated to get a very good model. Is it One or 10, or 1000 > or more?
That is highly dependent on the quality of your alignment. For example, if the sequence identity is very low, there is likely to be significant variability in the output models. (One way to see if you have adequate sampling is to plot the best molpdf for random samples of N models against N - if the plot does not flatten out with increasing N, more sampling is probably necessary.) In low sequence identity cases, you may also need to try slightly different alignments or perhaps entirely different templates, and some amount of loop modeling is possibly also necessary.
> If more than one model is generated, how to access model? What parameter > should be considered? Is it molpdf value, DOPE score or GA341 score? I > am always getting same value with GA341 score for my generated models > that is 1, but the molpdf value and DOPE score are conflicting. Please > suggest me which scoring method I should follow to select the best model.
In our own published benchmarks for model assessment methods, the DOPE score shows the best discriminatory power at separating good models from bad ones. But if GA341 is reporting all models have a score of 1, they are probably all good models. It is difficult to say which is "best" because proteins are not static structures anyway.
Ben Webb, Modeller Caretaker