Sphinx: merging knowledge-based andab initioapproaches to improve protein loop prediction

Abstract

MotivationLoops are often vital for protein function, however, their irregular structures make them difficult to model accurately. Current loop modelling algorithms can mostly be divided into two categories: knowledge-based, where databases of fragments are searched to find suitable conformations and ab initio, where conformations are generated computationally. Existing knowledge-based methods only use fragments that are the same length as the target, even though loops of slightly different lengths may adopt similar conformations. Here, we present a novel method, Sphinx, which combines ab initio techniques with the potential extra structural information contained within loops of a different length to improve structure prediction.ResultsWe show that Sphinx is able to generate high-accuracy predictions and decoy sets enriched with near-native loop conformations, performing better than the ab initio algorithm on which it is based. In addition, it is able to provide predictions for every target, unlike some knowledge-based methods. Sphinx can be used successfully for the difficult problem of antibody H3 prediction, outperforming RosettaAntibody, one of the leading H3-specific ab initio methods, both in accuracy and speed.Availability and ImplementationSphinx is available at http://opig.stats.ox.ac.uk/webapps/sphinx.Supplementary information Supplementary data are available at Bioinformatics online.

Simple Selection Procedure to Distinguish between Static and Flexible Loops

“…Ab initio modeling is based only on the physico-chemical information provided by the protein sequence. Among loop modeling software using homology modeling or Ab initio methods, a few are available as web servers, e.g., GalaxyLoopPS2 [17], DaReUs-Loop [18], LoopIng [19], Sphinx [20], and RCD+ [21], while some are available as standalone tools, such as MODELLER [22] and Rosetta-NGK (Next-generation KIC) [23]. A paper by DaReUs-Loop's developers gives a brief comparison of the accuracy of these methods [18].…”

DaReUS-Loop: accurate loop modeling using fragments from remote or unrelated proteins

“…Existing loop modeling methods can be divided into: ab initio based 30 – 35 , knowledge-based 36 – 38 and the combination of both methods 39 – 41 .…”

“…Flanks are regions before and after the loop to be modeled. For the completion of crystal structures, these methods are shown to generate successful results when similar fragments to the loop of interest exist in the database 41 . ArchPRED 42 considers the secondary structures flanking the missing loop, their relative orientation and the number of missing residues to identify candidate loop conformations.…”

“…Similar approaches are considered by others to predict complementary determining region (CDR) of antibodies 46 , 47 . Another recent method is Sphinx, which first performs data-based search to find fragments shorter than the loop of interest and obtains structural informations 41 . Then it applies ab initio methods to generate fragments of correct length.…”

“…Most of the existing loop modeling methods are shown to perform successful loop predictions in high-resolution crystal structures with accuracies of about 1-2 Å , if the loop is short (3–12 residues) 32 – 34 , 37 , 41 , 43 , 44 and increasing up to ~4 Å for larger sizes (≤20 amino acids) 37 , 41 , 43 , 44 . However, in practical applications, loops of interest are typically non-homologous regions of a homologous template.…”