Predicted binding site information improves model ranking in protein docking using experimental and computer-generated target structures
© Maheshwari and Brylinski. 2015
Received: 21 May 2015
Accepted: 30 October 2015
Published: 23 November 2015
Protein-protein interactions (PPIs) mediate the vast majority of biological processes, therefore, significant efforts have been directed to investigate PPIs to fully comprehend cellular functions. Predicting complex structures is critical to reveal molecular mechanisms by which proteins operate. Despite recent advances in the development of new methods to model macromolecular assemblies, most current methodologies are designed to work with experimentally determined protein structures. However, because only computer-generated models are available for a large number of proteins in a given genome, computational tools should tolerate structural inaccuracies in order to perform the genome-wide modeling of PPIs.
To address this problem, we developed eRankPPI, an algorithm for the identification of near-native conformations generated by protein docking using experimental structures as well as protein models. The scoring function implemented in eRankPPI employs multiple features including interface probability estimates calculated by eFindSitePPI and a novel contact-based symmetry score. In comparative benchmarks using representative datasets of homo- and hetero-complexes, we show that eRankPPI consistently outperforms state-of-the-art algorithms improving the success rate by ~10 %.
eRankPPI was designed to bridge the gap between the volume of sequence data, the evidence of binary interactions, and the atomic details of pharmacologically relevant protein complexes. Tolerating structure imperfections in computer-generated models opens up a possibility to conduct the exhaustive structure-based reconstruction of PPI networks across proteomes. The methods and datasets used in this study are available at www.brylinski.org/erankppi.
KeywordsProtein-protein interactions Protein docking Contact-based symmetry Protein models eRankPPI eFindSitePPI ZDOCK ZRANK
Most proteins work by interacting with other proteins to fulfill their molecular functions, therefore, quaternary assemblies are the key components of the vast majority of biological processes. Consequently, the structural characterization of protein-protein complexes provides valuable insights into protein function and association mechanisms, immensely contributing to the understanding of cellular interaction networks. The knowledge of atomic-level details of protein-protein interactions (PPIs) is required for a number of practical applications, for instance, it is critical for the design of therapeutics targeting protein interfaces [1, 2]. X-ray crystallography and NMR spectroscopy are the most widely used experimental techniques to determine protein complex structures. Nonetheless, these methods cannot keep pace with the rapidly growing number of protein interactions identified by high-throughput approaches such as yeast two-hybrid  and affinity purification techniques (co-immunoprecipitation , tandem affinity purification ) followed by mass spectrometry. The low stability of many complexes as well as significant efforts and high costs associated with experiments certainly impede the systems-level exploration of the molecular structures of protein assemblies. On that account, computational tools for PPI structure modeling bridge the gap between the volume of sequence data, the evidence of binary interactions, and the atomic details of pharmacologically relevant protein complexes.
Quaternary structure modeling to find the best relative orientation of monomers forming a stable complex can be performed using template-based or template-free techniques. Template-based methods use the similarity to known complex structures to model the interaction between a given pair target proteins. This strategy involves superposing target proteins onto the identified templates using either global or interfacial structure alignments . For instance, PRISM models quaternary structures by matching target proteins to a template interface selected from a representative database of the experimental structures of PPI complexes [7, 8]. In contrast, template-free approaches do not use any quaternary information from similar protein complexes; instead, these methods perform docking of the tertiary structures of receptor and ligand proteins. A typical docking calculation comprises two successive steps. First, a rigid-body sampling of six translational and rotational degrees of freedom generates a large set of candidate dimer conformations, in which the constituent monomers are in contact avoiding steric clashes. In the second step, a scoring function is used to rank the disparate collection of docked poses in order to identify near-native models. Current docking algorithms employ a variety of conformational search techniques including a fast Fourier transform [9–11], Monte Carlo methods , and the geometric hashing [13, 14]; for recent reviews see [15–17]. Significant efforts have also been devoted to develop reliable scoring functions, many of which assess the stability of the assembled dimers by combining multiple scoring terms such as the geometric shape [18–21], chemical and electrostatic complementarity [22–26]. Nevertheless, despite the advances in pose prediction and scoring, docking programs still face significant difficulties in identifying the best solution from a pool of candidates generated through conformational sampling [22, 27]. Therefore, the development of new approaches to more reliably distinguish between near-native and decoy conformations represents a practical strategy to improve the accuracy of protein docking.
To address the problem of model scoring, the prediction of protein quaternary structures is often supported by a variety of experimental and computational data [28–30]. Several strategies to incorporate experimental data in protein docking have been developed. For instance, upper bounds for distances between residues in interacting protein chains can be identified by NMR spectroscopy  and chemical crosslinking . Moreover, simultaneous screening for mutations that disrupt yeast two-hybrid interactions was proposed to identify critical interface residues for multiple interacting partners . Experimental data can be subsequently transformed into distance constrains to narrow the search space and to guide the selection of docking poses [34, 35]. Indeed, data-driven docking has been demonstrated to considerably improve the accuracy of dimer structure modeling , nonetheless, a limited availability of experimental data remains the major drawback of large-scale investigations of PPI networks. Although computational methods for interface residue prediction [37, 38] can support the complex assembly through PPI prediction-driven docking strategies, [38, 39] the predicted PPI site information is not always accurate leading to spurious results generated by a misguided conformational sampling.
Interaction symmetry is another commonly used form of constraints to model homo-oligomeric complexes. Symmetry is a prevalent feature of the global arrangement between subunits in homo-oligomer complexes formed by two or more identical protein chains. Homo-dimers are important parts of biochemical pathways that are found to occur more frequently than by chance . Approximately 50-70 % of the available datasets comprise homo-oligomers whose structural symmetry is remarkably well conserved [40–43]. The symmetric organization of proteins is known to confer structural and functional advantages providing stability, control over accessibility and specificity of active sites . It also provides the ability to avoid unwanted aggregation, which is responsible for a number of pathological conditions, such as Alzheimer’s and prion diseases [45, 46]. Furthermore, the symmetric self-association provides an opportunity for cooperative interactions and multivalent binding . Since the cyclic symmetry containing a single rotational axis is the most common type of regularity observed in protein quaternary structures, symmetrical docking a priori restricts the conformational search space only to symmetric transformations [10, 48].
In recent years, a two-stage ranking strategy has gained significant attention. Here, a standard protocol is first employed to rapidly scan for putative dimer conformations and to identify a subset of plausible candidates. Subsequently, an additional scoring system is used to re-rank the docked conformations in order to improve the ranking of near-native poses. These methods integrate a variety of features including sophisticated energy calculations, experimental and predicted binding site locations, statistical potentials derived from databases of complex structures, and evolutionary information [28, 49]. For instance, ZRANK  combines van der Waals, electrostatic and desolvation energy terms to re-rank the initial docking predictions generated by ZDOCK , whereas DECK  employs a distance and environment dependent knowledge-based potential to refine predictions from GRAMMX . Furthermore, the accuracy of HADDOCK  was improved by applying a scoring function based on a Voronoi tessellation of protein structures and machine learning . Other examples include T-PioDock , which uses interface prediction to assist the ranking of docked poses, and ClusPro  that re-ranks the top 2000 solutions generated either by ZDOCK or DOT  using a greedy clustering technique. Most of the available re-ranking protocols were designed and subsequently benchmarked using the experimentally determined structures in their bound and unbound conformational state. Since the structure-based reconstruction of across-proteome interaction networks involves docking of various quality homology models, re-ranking strategies should ideally tolerate inaccuracies in the atomic coordinates of interacting monomers.
In that regard, we developed eRankPPI, an algorithm for the selection of correct docking conformations constructed by protein docking using not only experimental monomer structures but also protein models. A scoring function implemented in eRankPPI combines in a novel way certain features such as residue-level interface probabilities estimated by eFindSitePPI , protein docking potentials , and a new contact-based symmetry score. Although, the predicted interface location was already successfully employed to improve the ranking accuracy for docked conformations , most previously reported benchmarking calculations were carried out against relatively small datasets of experimental structures [59–61]. In contrast, in this study, we perform a comprehensive analysis using non-redundant and representative sets of crystal structures as well as various quality protein models. In large-scale benchmarks using homo- and hetero-complexes, the accuracy of eFindSitePPI is compared to state-of-the-art scoring methods.
Datasets and tools
The algorithm for the re-ranking of docking models is trained and tested on the BM1905 dataset of 1905 proteins, which was compiled previously to evaluate the accuracy of interface residue prediction . This dataset contains experiment target structures (BM1905C) as well as high- and moderate- quality models (BM1905H and BM1905M, respectively). The quality of computer-generated models was assessed by TM-score , which ranges from 0 to 1 with values ≥0.4 indicating a significant structure similarity to the native protein. BM1905M and BM1905H datasets comprise models whose TM-score is in the range of 0.4–0.7 and 0.7–0.9 respectively. Furthermore, the BM1905 dataset contains 1755 homo-dimers (BM1755) and 150 hetero-dimers (BM150).
ZDOCK  version 3.0.2 is used to generate rigid-body docking conformations with the default search parameters. It has consistently been among the best performing algorithms in the Critical Assessment of Prediction of Interactions (CAPRI) [27, 63–66], a community-wide project assessing the accuracy of protein-protein docking algorithms. ZDOCK employs a fast Fourier transform (FFT) correlation-based method, which performs a systematic search in the six-dimensional space created by 3 rotational and 3 translational degrees of freedom. Docking conformations are predicted based on the desolvation and electrostatics contributions to the complex formation as well as the pairwise shape complementarity. Prior to docking, both the receptor and ligand structures are randomly translated and rotated to avoid any bias towards initial orientations. We collect 2000 highest scoring conformations reported by ZDOCK for each protein.
In this study, putative interfacial sites are predicted for the benchmarking receptors by eFindSitePPI , a recently developed structure/evolution-based approach to detect interface residues. eFindSitePPI exploits a general tendency of the location and geometry of binding sites to be highly conserved in evolutionarily weakly related dimer proteins. It employs a collection of effective algorithms, including meta-threading by eThread , structure alignments by Fr-TM-align , and machine learning using Support Vector Machines (SVMs) and a Naïve Bayes Classifier (NBC) . Each residue in the query protein is assigned a probability to be at the interface using residue-level attributes in combination with sequence and structure conservation scores derived from evolutionarily related templates.
eRankPPI developed in this study employs a series of attributes to re-rank docking conformations, including residue-level interface probabilities, protein docking contact potentials, and energy-based scores. The training and evaluation is performed separately for homo- and hetero-dimers as the modeling of homo-complex structures additionally takes account of symmetry constraints. Individual features are described below.
eRankPPI incorporates interface probability estimates for the receptor protein. We use probability scores assigned to each residue in the target protein by eFindSitePPI to estimate the likelihood to be at the protein-protein interface. Interfacial residues in docking models constructed by ZDOCK are identified by iAlign , which uses a distance-based criterion to identify the interface in a given multimer structure. The interface score is the sum of probabilities calculated over interface residues; two scores are computed using SVC and NBC. In general, these scores favor docking conformations with a substantial coverage of surface regions assigned a high interfacial probability by eFindSitePPI.
Protein docking potential
In addition to the interface scores, we employ a protein docking potential previously developed using a linear programming technique . In this model, the side chain center of mass, the backbone carbonyl oxygen, and the amide group are considered interaction sites for each residue. Inter-residue contacts are defined using distance thresholds of 6.8 Å, 4.0 Å and 5.6 Å for side chain, backbone and backbone/side chain sites, respectively. Two hundred fifty-three independent pairwise parameters were optimized in order to efficiently discriminate between hits and non-hits across protein-protein ensembles constructed by rigid-body docking.
ZDOCK energy score
Conformational ensembles of putative dimers are constructed by ZDOCK, as described above. The scoring function implemented in ZDOCK is a linear weighted sum of van der Waals attractive and repulsive energies, short- and long-range attractive and repulsive electrostatic energies, and desolvation. The optimal set of weight factors that maximizes the discriminatory capabilities of ZDOCK was obtained by training the scoring function on the Benchmark 1.0 set , followed by a cross-validation against non-homologous cases selected from the Benchmark 2.0 set . We use the total energy score reported by ZDOCK as one of the components of the scoring function in eRankPPI.
Essentially, the Jaccard index is a ratio of the intersection and the union between the two sets of interacting residue pairs, where Ax : By is considered a match for Ay : Bx. CBS ranges from 1 for perfectly symmetrical interfaces to 0 for completely asymmetrical complexes. For example, CBS scores calculated for homo-dimers shown in Fig. 1a and b, are 1 (perfect symmetry) and 1/3 (one-third of a perfect symmetry), respectively. The CSB scores are used only for homo-dimers, therefore five features are computed by eRankPPI for homo-complexes, whereas four features are used for hetero-dimers.
Evaluation of docking predictions
Assessment of model ranking
Protein docking algorithms typically construct multiple dimer models for a given pair of protein structures. Therefore, a reliable scoring function is critical to rank the predicted models so that near-native structures can be selected from a large set of decoys. In that regard, we evaluate the ranking capability using the following measures:
Percentage of successful cases
This metric reports the percentage of docking cases for which at least one hit is ranked within the top 10 models. Hits are defined as those conformations having iRMSD below a given cutoff varying from 0 to 15 Å. In addition to the iRMSD, we also calculate the percentage of successful cases using PCS as the hit criterion where the respective cutoff changes from 1 to 0.
Hit count gives the average number of hits within the top 10 docking models across the benchmarking dataset. Hits are predictions whose iRMSD is below a given cutoff ranging from 0 to 15 Å. Thus the hit count measures the overall enrichment of the top ranked models with near-native conformations.
The docking success rate is defined as the percentage of targets for which at least one correct model is ranked within the top n conformations, where n changes from 1 to 1000. The acceptance criteria for correct predictions are an iRMSD of ≤2.5 Å, ≤8.5 Å and ≤9.5 Å for experimental structures, high- and moderate-quality models, respectively.
Symmetry in homo-dimers
These findings encouraged us to use the CBS as one of the features to improve the ranking of homo-dimers. As a matter of fact, the concept of symmetry is widely used to construct homo-dimer complexes. Several protein docking programs were developed to model homo-oligomer structures by performing a systematic space search exclusively for symmetric conformations, e.g., M-ZDOCK , SymmRef  and SymmDock [48, 76]. These programs commonly use the symmetry to narrow the search space, however, eRankPPI employs a different approach. First, it incorporates the deviation from an ideal symmetry as a feature to improve the ranking of near-native models within docking ensembles generated through an unrestricted conformational search. Second, eRankPPI exploits a contacts-based symmetry rather than geometric regularities, which is more suitable for complex assembly using computer-generated monomers whose tertiary structures are somewhat distorted compared to experimental structures. To our knowledge, the pairwise contact-based symmetry is a novel feature used by eRankPPI in the modeling of homo-dimers.
Quality of predicted binding interfaces
The knowledge of PPI sites can be used to improve the success rate in protein docking [28, 36, 77]. Several groups integrated experimentally determined PPI information into their docking algorithms either to restrict the docking space during pose prediction or to filter the constructed conformations as a post-processing step. Moreover, due to the limited availability of experimental data, predicted PPI sites can be used instead. Nonetheless, the predicted PPI information is not always highly accurate and using erroneous data may lead to failed predictions. Ideally, docking strategies utilizing predicted PPI sites should tolerate to some extent only partially accurate constraints. In eRankPPI, we use interface residue prediction by eFindSitePPI that produces a continuous range of probability estimates over surface residues in target proteins rather than just a binary classification of interacting and non-interacting residues. These probability estimates are used to calculate the cumulative interface score for a given docking model, which is advantageous over the binary classification as it better tolerates a weaker signal from PPI prediction with moderate and low accuracy.
Ranking using experimental structures
Comparison of the success rates for different scoring functions against experimental target structures
Success rate [%]
iRMSD = 2.5 Å
PCS = 0.65
iRMSD = 2.5 Å
PCS = 0.65
Ranking using computer-generated models
Genome-wide determination of protein interaction networks is an important step in the elucidation of cellular regulatory mechanisms [80, 81]. Although constituent interactions can be modeled through a structure-based dimer assembly, the performance of scoring functions for model selection certainly depends on the quality on input structures. So far, we discussed the ranking of dimer models constructed from experimental monomer structures. Nonetheless, despite the exponential growth of the PDB, experimentally determined structures of a vast majority of gene products are not yet available. This necessitates using computer-generated models in protein docking, however, assuming that a docking program is capable to reliably construct complexes using theoretical monomer structures. Previously, a low-resolution docking method was applied to protein models  as a starting point for the subsequent high-resolution refinement to address the challenges of PPI modeling at a proteome-wide scale.
Here, we investigate how different docking scoring strategies cope with inaccuracies in the computer-generated models of query proteins. Undoubtedly, docking using protein models represents a difficult task and the quality of the resulting dimers cannot be higher than the quality of monomer structures. An iRMSD cutoff of 2.5 Å is widely accepted as a criterion for near-native models using experimental structures. However, different threshold values need to be used to evaluate dimer structures assembled from computer-generated models in order to account for distortions in individual monomers. Therefore, we first calculated the distribution of hits with an iRMSD of 2.5 Å across the top 2000 docking models constructed by ZDOCK using experimental monomer structures. A black dashed line in Fig. 5a shows that at least one assembled dimer has an iRMSD of 2.5 Å for about 70 % of the target proteins. We found that an iRMSD cutoff of 8.5 Å (9.5 Å) gives a similar coverage when high- (moderate-) quality models are used in protein docking. Furthermore, we established PCS cutoffs in a similar fashion so that ~70 % of the cases have at least one hit within docking ensembles; the corresponding threshold values are 0.65, 0.30 and 0.25 for crystal structures, high- and moderate-quality models, respectively.
Comparison of the success rates for different scoring functions against high- and moderate-quality protein models
Success rate [%]
iRMSD = 8.5 Å
PCS = 0.30
iRMSD = 9.5 Å
PCS = 0.25
Similar performance improvements are observed for the hit count and the success rate. For instance, Fig. 5g and k show that using eRankPPI yields an average number of 1.36 and 1.35 hits per target for the BM1755H and BM1755M datasets at the iRMSD cutoffs of 8.5 and 9.5 Å, respectively. For comparison, the corresponding hit counts for ZDOCK (ZRANK) are only 0.66 (0.69) and 0.46 (0.47). Furthermore, in Fig. 6, we examine the performance of eRankPPI, ZDOCK and ZRANK on the BM58H and BM58M datasets of hetero-dimers. For instance, Fig. 6b and c show that the percentage of successful cases at an iRMSD of 8.5 Å (9.5 Å) obtained by eRankPPI, ZDOCK and ZRANK for BM58H (BM58M) is 50.0 % (34.4 %), 29.31 % (27.5 %) and 34.5 % (17.2 %) respectively. This comprehensive analysis using various evaluation measures demonstrates that dimer ranking by eRankPPI is consistently better than that by ZDOCK and ZRANK not only using experimental monomer structures, but also computer-generated models.
The identification of near-native conformations across docking ensembles remains a challenging problem in the structure-based modeling of protein-protein interactions. Docking strategies need accurate scoring functions to rank the predicted conformations. Many current approaches employ the geometric, chemical and electrostatic complementarity as well as knowledge-based interaction potentials as components of their scoring functions. In this communication, we describe eRankPPI, a new scoring method for protein-protein docking that integrates predicted binding site information, protein docking potentials, energy-based scoring and a contact-based symmetry constraints (for homo-dimers). Although these attributes have been used previously in protein docking, we combined them in eRankPPI as a single, machine learning-based scoring function. The results demonstrate that eRankPPI reliably selects near-native conformations from a large number of decoys generated by ZDOCK . Moreover, comparative benchmarks show that eRankPPI consistently outperforms the state-of-the-art algorithms, ZDOCK and ZRANK, for both homo- and hetero-complexes yielding notably higher hit counts and success rates.
In addition to experimental target structures, we performed a series of benchmarking simulations using computer-generated models. Interestingly, ZRANK performs better than ZDOCK only against experimental target structures. The main reason for this high sensitivity to distortions in target structures is likely a strong dependence on atomic potentials, therefore, ZRANK requires high-quality structural data in order to provide accurate ranking. In contrast, eRankPPI outperforms both ZDOCK and ZRANK not only using experimental structures, but also computer-generated models. This is an important feature of eRankPPI owing to the fact that protein models represent the most challenging targets for molecular docking.
The analysis of the linear regression model used by eRankPPI to rank hetero-dimers shows that the optimized weights for the SVC and NBC interface scores assigned by eFindSitePPI, the protein-docking potential and the ZDOCK score are 171.9, 891.8, 122.7 and 2.2, respectively. Therefore, the predicted binding site information is a major contributor to the improvement of model ranking in protein docking. Since the success of eRankPPI depends on the accuracy of protein interface prediction, using a robust PPI prediction program is essential. Here, we use eFindSitePPI, a recently developed template-based approach that effectively exploits the tendency of the location of binding sites to be highly conserved across evolutionarily related protein dimers . eFindSitePPI uses the three-dimensional structure of a query protein, evolutionarily remotely related templates and machine learning to predict interfacial sites. It was also shown to outperform several PPI site prediction programs . Also, different from other prediction techniques, eFindSitePPI tolerates structural imperfections in computer-generated models. These characteristics make eFindSitePPI a preferred PPI predictor to support dimer ranking in across-proteome docking studies using eRankPPI.
In this study, we developed eRankPPI, an algorithm for the selection of correct docking conformations constructed by rigid-body protein docking. eRankPPI features a new scoring function that integrates the predicted interface location with protein docking potentials and a contact-based symmetry score. Comprehensive benchmarking calculations show that eRankPPI has a high tolerance to structural imperfections in computer-generated protein models, therefore, it opens up a possibility to conduct the exhaustive structure-based reconstruction of PPI networks across proteomes.
Availability of supporting data
The methods and datasets used in this study are available at www.brylinski.org/erankppi.
Contact-based symmetry score
Matthew’s correlation coefficient
MCC-based pairwise contact score
Protein data bank
Support vector machines
This study was supported by the Louisiana Board of Regents through the Board of Regents Support Fund [contract LEQSF (2012–15)-RD-A-05].
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Berg T. Modulation of protein-protein interactions with small organic molecules. Angew Chem Int Ed Engl. 2003;42:2462–81.View ArticlePubMedGoogle Scholar
- Meireles LMC, Mustata G. Discovery of modulators of protein-protein interactions: current approaches and limitations. Curr Top Med Chem. 2011;11:248–57.View ArticlePubMedGoogle Scholar
- Fields S, Song O. A novel genetic system to detect protein-protein interactions. Nature. 1989;340:245–6.View ArticlePubMedGoogle Scholar
- Ho Y, Gruhler A, Heilbut A, Bader GD, Moore L, Adams S-L, et al. Systematic identification of protein complexes in Saccharomyces cerevisiae by mass spectrometry. Nature. 2002;415:180–3.View ArticlePubMedGoogle Scholar
- Rigaut G, Shevchenko A, Rutz B, Wilm M, Mann M, Séraphin B. A generic protein purification method for protein complex characterization and proteome exploration. Nat Biotechnol. 1999;17:1030–2.View ArticlePubMedGoogle Scholar
- Tuncbag N, Gursoy A, Keskin O. Prediction of protein-protein interactions: unifying evolution and structure at protein interfaces. Phys Biol. 2011;8:035006.View ArticlePubMedGoogle Scholar
- Tuncbag N, Keskin O, Nussinov R, Gursoy A. Fast and accurate modeling of protein-protein interactions by combining template-interface-based docking with flexible refinement. Proteins. 2012;80:1239–49.View ArticlePubMedGoogle Scholar
- Baspinar A, Cukuroglu E, Nussinov R, Keskin O, Gursoy A. PRISM: a web server and repository for prediction of protein-protein interactions and modeling their 3D complexes. Nucleic Acids Res. 2014;42(Web Server issue):W285–9.PubMed CentralView ArticlePubMedGoogle Scholar
- Chen R, Li L, Weng Z. ZDOCK: an initial-stage protein-docking algorithm. Protiens. 2003;52(1):80–7. November 2002.View ArticleGoogle Scholar
- Pierce B, Tong W, Weng Z. M-ZDOCK: A grid-based approach for Cn symmetric multimer docking. Bioinformatics. 2005;21:1472–8.View ArticlePubMedGoogle Scholar
- Sinha R, Kundrotas PJ, Vakser IA. Docking by structural similarity at protein-protein interfaces. Proteins. 2010;78:3235–41.PubMed CentralView ArticlePubMedGoogle Scholar
- Gray JJ, Moughon S, Wang C, Schueler-Furman O, Kuhlman B, Rohl CA, et al. Protein-protein docking with simultaneous optimization of rigid-body displacement and side-chain conformations. J Mol Biol. 2003;331:281–99.View ArticlePubMedGoogle Scholar
- Venkatraman V, Yang YD, Sael L, Kihara D. Protein-protein docking using region-based 3D Zernike descriptors. BMC Bioinformatics. 2009;10:407.PubMed CentralView ArticlePubMedGoogle Scholar
- Fischer D, Lin SL, Wolfson HL, Nussinov R. A geometry-based suite of molecular docking processes. J Mol Biol. 1995;248:459–77.PubMedGoogle Scholar
- Moal IH, Moretti R, Baker D, Fernández-Recio J. Scoring functions for protein-protein interactions. Curr Opin Struct Biol. 2013;23:862–7.View ArticlePubMedGoogle Scholar
- Huang S-Y. Search strategies and evaluation in protein-protein docking: principles, advances and challenges. Drug Discov Today. 2014;19:1081–96.View ArticlePubMedGoogle Scholar
- Huang S-Y. Exploring the potential of global protein-protein docking: an overview and critical assessment of current programs for automatic ab initio docking. Drug Discov Today. 2015;20:969–77.View ArticlePubMedGoogle Scholar
- Chang S, Jiao X, Li C, Gong X, Chen W, Wang C. Amino acid network and its scoring application in protein-protein docking. Biophys Chem. 2008;134:111–8.View ArticlePubMedGoogle Scholar
- Khashan R, Zheng W, Tropsha A. Scoring protein interaction decoys using exposed residues (SPIDER): a novel multibody interaction scoring function based on frequent geometric patterns of interfacial residues. Proteins. 2012;80:2207–17.PubMed CentralView ArticlePubMedGoogle Scholar
- Mitra P, Pal D. New measures for estimating surface complementarity and packing at protein-protein interfaces. FEBS Lett. 2010;584:1163–8.View ArticlePubMedGoogle Scholar
- Pons C, Glaser F, Fernandez-Recio J. Prediction of protein-binding areas by small-world residue networks and application to docking. BMC Bioinformatics. 2011;12:378.PubMed CentralView ArticlePubMedGoogle Scholar
- Halperin I, Ma B, Wolfson H, Nussinov R. Principles of docking: An overview of search algorithms and a guide to scoring functions. Proteins. 2002;47:409–43.View ArticlePubMedGoogle Scholar
- Andrusier N, Mashiach E, Nussinov R, Wolfson HJ. Principles of flexible protein-protein docking. Proteins. 2008;73(2):271–89.PubMed CentralView ArticlePubMedGoogle Scholar
- Demir-Kavuk O, Krull F, Chae M-H, Knapp E-W. Predicting protein complex geometries with linear scoring functions. Genome Inform. 2010;24:21–30.PubMedGoogle Scholar
- Cheng TM-K, Blundell TL, Fernandez-Recio J. pyDock: electrostatics and desolvation for effective scoring of rigid-body protein-protein docking. Proteins. 2007;68:503–15.View ArticlePubMedGoogle Scholar
- Lyskov S, Gray JJ. The RosettaDock server for local protein-protein docking. Nucleic Acids Res. 2008;36(Web Server issue):W233–8.PubMed CentralView ArticlePubMedGoogle Scholar
- Lensink MF, Wodak SJ. Docking and scoring protein interactions: CAPRI 2009. Proteins. 2010;78:3073–84.View ArticlePubMedGoogle Scholar
- Alber F, Förster F, Korkin D, Topf M, Sali A. Integrating diverse data for structure determination of macromolecular assemblies. Annu Rev Biochem. 2008;77:443–77.View ArticlePubMedGoogle Scholar
- de Vries SJ, van Dijk M, Bonvin AMJJ. The HADDOCK web server for data-driven biomolecular docking. Nat Protoc. 2010;5:883–97.View ArticlePubMedGoogle Scholar
- Huang B, Schroeder M. Using protein binding site prediction to improve protein docking. Gene. 2008;422:14–21.View ArticlePubMedGoogle Scholar
- van Dijk ADJ, Fushman D, Bonvin AMJJ. Various strategies of using residual dipolar couplings in NMR-driven protein docking: application to Lys48-linked di-ubiquitin and validation against 15 N-relaxation data. Proteins. 2005;60:367–81.View ArticlePubMedGoogle Scholar
- Meenan NAG, Sharma A, Fleishman SJ, Macdonald CJ, Morel B, Boetzel R, et al. The structural and energetic basis for high selectivity in a high-affinity protein-protein interaction. Proc Natl Acad Sci U S A. 2010;107:10080–5.PubMed CentralView ArticlePubMedGoogle Scholar
- Hill RB, Manlandro CM. Two-hybrid based screen to identify disruptive residues at multiple protein interfaces. 2012. U.S. Patent No 20120157323 A1.Google Scholar
- Shih ESC, Hwang MJ. On the use of distance constraints in protein-protein docking computations. Proteins. 2012;80:194–205.View ArticlePubMedGoogle Scholar
- Karaca E, Melquiond ASJ, de Vries SJ, Kastritis PL, Bonvin AMJJ. Building macromolecular assemblies by information-driven docking: introducing the HADDOCK multibody docking server. Mol Cell Proteomics. 2010;9:1784–94.PubMed CentralView ArticlePubMedGoogle Scholar
- Shih ESC, Hwang M-J. A critical assessment of information-guided protein-protein docking predictions. Mol Cell Proteomics. 2013;12:679–86.PubMed CentralView ArticlePubMedGoogle Scholar
- Sites PI, Porollo A, Meller J. Computational methods for prediction of protein-protein interaction sites. InTech. 2008; doi: 10.5772/36716.
- de Vries SJ, Bonvin AMJJ. CPORT: a consensus interface predictor and its performance in prediction-driven docking with HADDOCK. PLoS One. 2011;6:e17695.PubMed CentralView ArticlePubMedGoogle Scholar
- Huang S-Y, Zou X. An iterative knowledge-based scoring function for protein-protein recognition. Proteins. 2008;72:557–79.View ArticlePubMedGoogle Scholar
- Ispolatov I, Yuryev A, Mazo I, Maslov S. Binding properties and evolution of homodimers in protein-protein interaction networks. Nucleic Acids Res. 2005;33:3629–35.PubMed CentralView ArticlePubMedGoogle Scholar
- Levy ED, Pereira-Leal JB, Chothia C, Teichmann SA. 3D complex: a structural classification of protein complexes. PLoS Comput Biol. 2006;2:e155.PubMed CentralView ArticlePubMedGoogle Scholar
- Levy ED, Boeri Erba E, Robinson CV, Teichmann SA. Assembly reflects evolution of protein complexes. Nature. 2008;453:1262–5.PubMed CentralView ArticlePubMedGoogle Scholar
- Dayhoff JE, Shoemaker BA, Bryant SH, Panchenko AR. Evolution of protein binding modes in homooligomers. J Mol Biol. 2010;395:860–70.PubMed CentralView ArticlePubMedGoogle Scholar
- Marianayagam NJ, Sunde M, Matthews JM. The power of two: protein dimerization in biology. Trends Biochem Sci. 2004;29:618–25.View ArticlePubMedGoogle Scholar
- Hayouka Z, Rosenbluh J, Levin A, Loya S, Lebendiker M, Veprintsev D, et al. Inhibiting HIV-1 integrase by shifting its oligomerization equilibrium. Proc Natl Acad Sci U S A. 2007;104:8316–21.PubMed CentralView ArticlePubMedGoogle Scholar
- Wright CF, Teichmann SA, Clarke J, Dobson CM. The importance of sequence diversity in the aggregation and evolution of proteins. Nature. 2005;438:878–81.View ArticlePubMedGoogle Scholar
- Goodsell D, Olson A. Structural symmetry and protein function. Annu Rev Biophys Biomol Struct. 2000;29:105–53.View ArticlePubMedGoogle Scholar
- Schneidman-Duhovny D, Inbar Y, Nussinov R, Wolfson HJ. PatchDock and SymmDock: servers for rigid and symmetric docking. Nucleic Acids Res. 2005;33(Web Server issue):W363–7.PubMed CentralView ArticlePubMedGoogle Scholar
- Ritchie DW. Recent progress and future directions in protein-protein docking. Curr Protein Pept Sci. 2008;9:1–15.View ArticlePubMedGoogle Scholar
- Pierce B, Weng Z. ZRANK: reranking protein docking predictions with an optimized energy function. Proteins. 2007;67(4):1078–86. October 2006.View ArticlePubMedGoogle Scholar
- Liu S, Vakser IA. DECK: Distance and environment-dependent, coarse-grained, knowledge-based potentials for protein-protein docking. BMC Bioinformatics. 2011;12:280.PubMed CentralView ArticlePubMedGoogle Scholar
- Tovchigrechko A, Vakser IA. GRAMM-X public web server for protein-protein docking. Nucleic Acids Res. 2006;34(Web Server issue):W310–4.PubMed CentralView ArticlePubMedGoogle Scholar
- Bernauer J, Azé J, Janin J, Poupon A. A new protein-protein docking scoring function based on interface residue properties. Bioinformatics. 2007;23:555–62.View ArticlePubMedGoogle Scholar
- Esmaielbeiki R, Nebel J-C. Scoring docking conformations using predicted protein interfaces. BMC Bioinformatics. 2014;15:171.PubMed CentralView ArticlePubMedGoogle Scholar
- Comeau SR, Gatchell DW, Vajda S, Camacho CJ. ClusPro: A fully automated algorithm for protein-protein docking. Nucleic Acids Res. 2004;32:96–9.View ArticleGoogle Scholar
- Roberts VA, Thompson EE, Pique ME, Perez MS, Ten Eyck LF. DOT2: Macromolecular docking with improved biophysical models. J Comput Chem. 2013;34:1743–58.PubMed CentralView ArticlePubMedGoogle Scholar
- Maheshwari S, Brylinski M. Prediction of protein-protein interaction sites from weakly homologous template structures using meta-threading and machine learning. J Mol Recognit. 2015;28:35–48.View ArticlePubMedGoogle Scholar
- Tobi D, Bahar I. Optimal design of protein docking potentials: efficiency and limitations. Proteins. 2006;62:970–81.View ArticlePubMedGoogle Scholar
- Hwang H, Pierce B, Mintseris J, Joel Janin ZW. Protein-protein docking benchmark version 3.0. Proteins. 2009;73:705–9.View ArticleGoogle Scholar
- Hwang H, Vreven T, Janin J, Weng Z. Protein-protein docking benchmark version 4.0. Proteins. 2010;78:3111–4.PubMed CentralView ArticlePubMedGoogle Scholar
- Janin J, Wodak S. The third CAPRI assessment meeting Toronto, Canada, April 20–21, 2007. Structure. 2007;15:755–9.View ArticlePubMedGoogle Scholar
- Pandit SB, Skolnick J. Fr-TM-align: a new protein structural alignment method based on fragment alignments and the TM-score. BMC Bioinformatics. 2008;9:531.PubMed CentralView ArticlePubMedGoogle Scholar
- Lensink MF, Wodak SJ. Blind predictions of protein interfaces by docking calculations in CAPRI. Proteins. 2010;78:3085–95.View ArticlePubMedGoogle Scholar
- Chen R, Tong W, Mintseris J, Li L, Weng Z. ZDOCK predictions for the CAPRI challenge. Proteins. 2003;52:68–73.View ArticlePubMedGoogle Scholar
- Hwang H, Vreven T, Pierce BG, Hung JH, Weng Z. Performance of ZDOCK and ZRANK in CAPRI rounds 13–19. Proteins. 2010;78:3104–10.PubMed CentralView ArticlePubMedGoogle Scholar
- Wiehe K, Pierce B, Wei WT, Hwang H, Mintseris J, Weng Z. The performance of ZDOCK and ZRANK in rounds 6–11 of CAPRI. Proteins. 2007;69:719–25.View ArticlePubMedGoogle Scholar
- Brylinski M, Lingam D. eThread: a highly optimized machine learning-based approach to meta-threading and the modeling of protein tertiary structures. PLoS One. 2012;7:e50200.PubMed CentralView ArticlePubMedGoogle Scholar
- Zhang H. The optimality of naive bayes. Mach Learn. 2004;1:3.Google Scholar
- Gao M, Skolnick J. iAlign: a method for the structural comparison of protein-protein interfaces. Bioinformatics. 2010;26:2259–65.PubMed CentralView ArticlePubMedGoogle Scholar
- Chen R, Mintseris J, Janin J, Weng Z. A protein-protein docking benchmark. Proteins. 2003;52:88–91.View ArticlePubMedGoogle Scholar
- Mintseris J, Wiehe K, Pierce B, Anderson R, Chen R, Janin J, et al. Protein-protein docking benchmark 2.0: an update. Proteins. 2005;60:214–6.View ArticlePubMedGoogle Scholar
- Chang C-C, Lin C-J. LIBSVM: a library for support vector machines. ACM Trans Intell Syst Technol. 2011;2:1–39.View ArticleGoogle Scholar
- Gentleman WM, University of Waterloo. Basic description for large, sparse or weighted linear least squares problems (Algorithm AS 75). Appl Stat. 1974;23:448–54.View ArticleGoogle Scholar
- Méndez R, Leplae R, Lensink MF, Wodak SJ. Assessment of CAPRI predictions in rounds 3–5 shows progress in docking procedures. Proteins. 2005;60:150–69.View ArticlePubMedGoogle Scholar
- Mashiach E, Nussinov R, Wolfson HJ. SymmRef: a flexible refinement method for symmetric multimers. Proteins. 2012;29:997–1003.Google Scholar
- Schneidman-Duhovny D, Inbar Y, Nussinov R, Wolfson HJ. Geometry-based flexible and symmetric protein docking. Proteins. 2005;60(January):224–31.View ArticlePubMedGoogle Scholar
- Li L, Huang Y, Xiao Y. How to use not-always-reliable binding site information in protein-protein docking prediction. PLoS One. 2013;8:e75936. doi:10.1371/journal.pone.0075936.PubMed CentralView ArticlePubMedGoogle Scholar
- Lichtarge O, Bourne HR, Cohen FE. An evolutionary trace method defines binding surfaces common to protein families. J Mol Biol. 1996;257:342–58.View ArticlePubMedGoogle Scholar
- Engelen S, Trojan LA, Sacquin-Mora S, Lavery R, Carbone A. Joint evolutionary trees: a large-scale method to predict protein interfaces based on sequence sampling. PLoS Comput Biol. 2009;5:e1000267.PubMed CentralView ArticlePubMedGoogle Scholar
- Bonetta L. Protein-protein interactions: Interactome under construction. Nature. 2010;468:851–4.View ArticlePubMedGoogle Scholar
- Vidal M, Cusick ME, Barabási A-L. Interactome networks and human disease. Cell. 2011;144:986–98.PubMed CentralView ArticlePubMedGoogle Scholar
- Tovchigrechko A, Wells CA, Vakser IA. Docking of protein models. Protein Sci. 2002;11:1888–96.PubMed CentralView ArticlePubMedGoogle Scholar
- Maheshwari S, Brylinski M. Predicting protein interface residues using easily accessible on-line resources. Brief Bioinform. 2015; doi: 10.1093/bib/bbv009.
- Okamoto A, Nakai Y, Hayashi H, Hirotsu K, Kagamiyama H. Crystal structures of Paracoccus denitrificans aromatic amino acid aminotransferase: a substrate recognition site constructed by rearrangement of hydrogen bond network. J Mol Biol. 1998;280:443–61.View ArticlePubMedGoogle Scholar
- Bell CE, Frescura P, Hochschild A, Lewis M. Crystal structure of the lambda repressor C-terminal domain provides a model for cooperative operator binding. Cell. 2000;101:801–11.View ArticlePubMedGoogle Scholar
- Bourne Y, Watson MH, Hickey MJ, Holmes W, Rocque W, Reed SI, et al. Crystal structure and mutational analysis of the human CDK2 kinase complex with cell cycle-regulatory protein CksHs1. Cell. 1996;84:863–74.View ArticlePubMedGoogle Scholar