The SIFT Server: General Information
The SIFT server can be used for ordering putative models of protein structures and filtering out non-native structure with non-physical packing in order to enhance protein folding simulations and protein structure prediction. Our filters were primarily designed to be used in the context of a large number of putative models in order to improve the convergence of folding simulation methods, such as Rosetta. However, individual models may be assessed independently as well (see MQAP competition in CAFASP).
Several sequence independent and sequence dependent filters (despite the name of the server :-) are available at present. The following sequence independent filters can be used to filter out structures with non-physical packing: the number of contacts, contact order, and a novel filter based on the shape of interresidue radial distribution function (referred to as the RDF shape filter). The essence of RDF shape filter, which is described in a paper available below, is the use of averaged (i.e. amino acid independent) RDF to discriminate properly packed models from misfolded ones. The current version of the SIFT server produces optionally several alternative scores. The sequence dependent filters include information about predicted relative solvent accessibility (RSA) and order the models according to the level of agreement between the predicted (using our SABLE server) and observed RSAs. The set of structures submitted to the server (either in PDB or LOOPP format; all structures zipped into one file, except for single models that can be submitted directly in PDB format) will be ranked according to the filter that you chose. The number of non local inter-residue contacts (contacts between sites separated by fewer than six virtual bonds are excluded) and the pairwise contact energy (according to the Myazawa-Jernigan potential) are given in each case. The optional output includes either contact order or inter-residue radial distribution function shape measure for each structure. Low energy, high contact order and high RDF shape measure may indicate strutcures with correct packing.
Sequence independent filters are expected to work for soluble globular proteins only. If you suspect that your protein may have an "open" conformation (you may use our SABLE server to predict if your structure has a hydrophobic core) you should not rely on the ranking of structures provided by SIFT. Moreover, SIFT is meant to be used for de novo simulations and alignment based models should not be evaluated using it. At the moment we only have a version for models with side-chains and the CA-only models will be effectively ignored (this also concerns models too short with respect to target sequences and models with only few contacts).
Its stand-alone version is also available. Click here to go to download page.
Adamczak R., Meller J. On the Transferability of Folding and Threading Potentials and Sequence-Independent Filters for Protein Folding Simulations Molecular Physics, vol. 102 (11-12): 1291-1305 (2004).