The CAPRI Evaluation Procedure

On the basis of a consensus reached with the CAPRI community, a total of eight parameters are evaluated. Two assess residue-residue contacts between the docked proteins. f nat is the fraction of intersubunit residue-residue contacts present in the X-ray structure of the target that are reproduced (recalled) in the model (residues are deemed in contact if any of their atoms are within 5 A) f non-nat is the fraction of residue contacts in 

It was realized early on that the fraction of native contacts f nat can be made artificially high if the subunits interpenetrate, and it was decided that models with more than a threshold number of atomic clashes (atom pairs separated by less than 3 A) should be disqualified. This threshold is derived from the distribution of the number of clashes observed in the ensemble of submitted models and equal to the average number of clashes plus twice the standard deviation of this distribution. Typically, 40-150 clashes are allowed in a model, depending on the size of the proteins and the interfaces. 

The extent to which a model identifies the correct regions of the protein surfaces that interact may be assessed independently of the contacts between these regions. For this purpose, two parameters f IR and f OP are computed that represent respectively the fraction of interface residues in the target that is reproduced in the model and the fraction of interface residues in the model that does not match those in the target. Both are evaluated separately for each component of the complex. Interface residues are defined as those that lose accessible surface area when the two proteins associate, computed from the difference in solvent accessible surface area of the complex and the individual components taken in the bound form. 

The final ranking of the models submitted for each target is based on a subset of these criteria, namely, f nat, L rms, and I rms. Their values define a model as being either incorrect, or of acceptable, medium, or high quality. A high-quality model must have f nat above 0.5, while both L rms and I rms are below 1.0 A, which in practice requires that it approach the quality of an 

To ensure a fair and consistent evaluation, all the assessment parameters must be calculated on the same number of residues and atoms across all submitted models. The submissions are therefore filtered to exclude flexible residues that adopt different conformations in the bound and unbound protein components, or not present across all models. These residues are defined using structural alignments of the individual ligand and receptor entities and identifying those with largest deviations. Typically, residues located in (short) loops are excluded from the RMSD fits and subsequent analysis. 

---