The cited work (Melo et.al., 2002) discusses the derivation of the statistical potentials used in the composite score. Please use "B. John, A. Sali. Comparative Protein Structure Modeling by Iterative Alignment, Model Building, and Model Assessment. Nucleic Acids Research 31, 3982-3992, 2003" to understand how the GA341 score works.
FYI:
GA341 = 1 - [ cos(sequence_identity) ]^(compactness +sequence_identity)/exp(z-score)
Sequence identity is the fraction of positions with identical residues in the target- template alignment. Structural compactness is the ratio between the sum of the standard volumes of the amino acid residues in the protein and the volume of the sphere with the diameter equal to the largest dimension of the model. The Z-score is calculated for the combined statistical potential energy of a model, using the mean and standard deviation of the 200 random sequences with the same composi- tion and structure as the model (Melo et al., 2002). The combined statistical poten- tial energy of a model is the sum of the solvent accessibility terms for all Cβ atoms and distance-dependent terms for all pairs of Cα and Cβ atoms. The solvent acces- sibility term for a Cβ atom depends on its residue type and the number of other Cβ atoms within 10Å; the non-bonded terms depend on the atom and residue types spanning the distance, the distance itself, and the number of residues separating the distance-spanning atoms in sequence. These potential terms reflect the statisti- cal preferences observed in 760 non-redundant proteins of known structure. The GA341 scoring function was evolved by a genetic algorithm that explored many combinations of a variety of mathematical functions and model features, to opti- mize the discrimination between good and bad models in a training set of models. The GA341 score ranges from 0 for models that tend to have an incorrect fold to 1 for models that tend to be comparable to at least low-resolution x- ray structures. GA341 scores greater than 0.7 indicate a correct fold with more than 35% of the backbone atoms superposable to better than 3.5Å.
Eswar.
--- Eswar Narayanan University of California - San Francisco
On Jul 8, 2006, at 7:45 AM, Cassiano wrote:
> Hi all, > > I have some questions about the model.asses_ga341. In the manual on- > line > is described that it uses the "percentage sequence identity between > the > template and the model as a parameter". But, in the cited work > (Melo et > al, 2002) i don't see this relation with sequence identity. My > questions: > - How Modeller treat the sequence identity in the context of the > cited work? > - If the identity between my target and templates is very low, how the > quality of the model is affected? (and what you recommend to use to > assess the quality) > > Thanks, > Cassiano Carromeu. > _______________________________________________ > modeller_usage mailing list > modeller_usage@salilab.org > http://salilab.org/mailman/listinfo/modeller_usage