Integrating atom-based and residue-based scoring functions for protein-protein docking

Most scoring functions for protein-protein docking algorithms are either atom-based or residue-based, with the former being able to produce higher quality structures and latter more tolerant to conformational changes upon binding. Earlier, we developed the ZRANK algorithm for reranking docking predictions, with a scoring function that contained only atom-based terms. Here we combine ZRANK's atom-based potentials with five residue-based potentials published by other labs, as well as an atom-based potential IFACE that we published after ZRANK. We simultaneously optimized the weights for selected combinations of terms in the scoring function, using decoys generated with the protein-protein docking algorithm ZDOCK. We performed rigorous cross validation of the combinations using 96 test cases from a docking benchmark. Judged by the integrative success rate of making 1000 predictions per complex, addition of IFACE and the best residue-based pair potential reduced the number of cases without a correct prediction by 38 and 27% relative to ZDOCK and ZRANK, respectively. Thus combination of residue-based and atom-based potentials into a scoring function can improve performance for protein-protein docking. The resulting scoring function is called IRAD (integration of residue- and atom-based potentials for docking) and is available at http://zlab.umassmed.edu.     

CyClus: A fast,comprehensive cylindrical interface approximation clustering/reranking method for rigid‐body protein–protein docking decoys

We propose a fast clustering and reranking method, CyClus, for protein–protein docking decoys. This method enables comprehensive clustering of whole decoys generated by rigid‐body docking using cylindrical approximation of the protein–proteininterface and hierarchical clustering procedures. We demonstrate the clustering and reranking of 54,000 decoy structures generated by ZDOCK for each complex within a few minutes. After parameter tuning for the test set in ZDOCK benchmark 2.0 with the ZDOCK and ZRANK scoring functions, blind tests for the incremental data in ZDOCK benchmark 3.0 and 4.0 were conducted. CyClus successfully generated smaller subsets of decoys containing near‐native decoys. For example, the number of decoys required to create subsets containing near‐native decoys with 80% probability was reduced from 22% to 50% of the number required in the original ZDOCK. Although specific ZDOCK and ZRANK results were demonstrated, the CyClus algorithm was designed to be more general and can be applied to a wide range of decoys and scoring functions by adjusting just two parameters, p and T. CyClus results were also compared to those from ClusPro. Proteins 2013; © 2012 Wiley Periodicals, Inc.     

The performance of ZDOCK and ZRANK in rounds 6-11 of CAPRI

We present an evaluation of our protein-protein docking approach using the ZDOCK and ZRANK algorithms, in combination with structural clustering and filtering, utilizing biological data in Rounds 6-11 of the CAPRI docking experiment. We achieved at least one prediction of acceptable accuracy for five of six targets submitted. In addition, two targets resulted in medium-accuracy predictions. In the new scoring portion of the CAPRI exercise, we were able to attain at least one acceptable prediction for the three targets submitted and achieved three medium-accuracy predictions for Target 26. Scoring was performed using ZRANK, a new algorithm for reranking initial-stage docking predictions using a weighted energy function and no structural refinement. Here we outline a practical and successful docking strategy, given limited prior biological knowledge of the complex to be predicted.     

Predicting protein complex geometries with a neural network

A major challenge of the protein docking problem is to define scoring functions that can distinguish near‐native protein complex geometries from a large number of non‐native geometries (decoys) generated with noncomplexed protein structures (unbound docking). In this study, we have constructed a neural network that employs the information from atom‐pair distance distributions of a large number of decoys to predict protein complex geometries. We found that docking prediction can be significantly improved using two different types of polar hydrogen atoms. To train the neural network, 2000 near‐native decoys of even distance distribution were used for each of the 185 considered protein complexes. The neural network normalizes the information from different protein complexes using an additional protein complex identity input neuron for each complex. The parameters of the neural network were determined such that they mimic a scoring funnel in the neighborhood of the native complex structure. The neural network approach avoids the reference state problem, which occurs in deriving knowledge‐based energy functions for scoring. We show that a distance‐dependent atom pair potential performs much better than a simple atom‐pair contact potential. We have compared the performance of our scoring function with other empirical and knowledge‐based scoring functions such as ZDOCK 3.0, ZRANK, ITScore‐PP, EMPIRE, and RosettaDock. In spite of the simplicity of the method and its functional form, our neural network‐based scoring function achieves a reasonable performance in rigid‐body unbound docking of proteins. Proteins 2010. © 2009 Wiley‐Liss, Inc.     

Efficient comprehensive scoring of docked protein complexes using probabilistic support vector machines

Biological systems and processes rely on a complex network of molecular interactions. While the association of biological macromolecules is a fundamental biochemical phenomenon crucial for the understanding of complex living systems, protein-protein docking methods aim for the computational prediction of protein complexes from individual subunits. Docking algorithms generally produce large numbers of putative protein complexes with only few of these conformations resembling the native complex structure within an acceptable degree of structural similarity. A major challenge in the field of docking is to extract near-native structure(s) out of the large pool of solutions, the so called scoring or ranking problem. A series of structural, chemical, biological and physical properties are used in this work to classify docked protein-protein complexes. These properties include specialized energy functions, evolutionary relationship, class specific residue interface propensities, gap volume, buried surface area, empiric pair potentials on residue and atom level as well as measures for the tightness of fit. Efficient comprehensive scoring functions have been developed using probabilistic Support Vector Machines in combination with this array of properties on the largest currently available protein-protein docking benchmark. The established classifiers are shown to be specific for certain types of protein-protein complexes and are able to detect near-native complex conformations from large sets of decoys with high sensitivity. Using classification probabilities the ranking of near-native structures was drastically improved, leading to a significant enrichment of near-native complex conformations within the top ranks. It could be shown that the developed schemes outperform five other previously published scoring functions.     

Protein-protein docking in CAPRI using ATTRACT to account for global and local flexibility

A reduced protein model combined with a systematic docking approach has been employed to predict protein-protein complex structures in CAPRI rounds 6-11. The docking approach termed ATTRACT is based on energy minimization in translational and rotational degrees of freedom of one protein with respect to the second protein starting from many thousand initial protein partner placements. It also allows for approximate inclusion of global flexibility of protein partners during systematic docking by conformational relaxation of the partner proteins in precalculated soft collective backbone degrees of freedom. We have submitted models for six targets, achieved acceptable docking solutions for two targets, and predicted >20% correct contacts for five targets. Possible improvements of the docking approach in particular at the scoring and refinement steps are discussed.     

Scoring optimisation of unbound protein-protein docking including protein binding site predictions

The prediction of the structure of the protein-protein complex is of great importance to better understand molecular recognition processes. During systematic protein-protein docking, the surface of a protein molecule is scanned for putative binding sites of a partner protein. The possibility to include external data based on either experiments or bioinformatic predictions on putative binding sites during docking has been systematically explored. The external data were included during docking with a coarse-grained protein model and on the basis of force field weights to bias the docking search towards a predicted or known binding region. The approach was tested on a large set of protein partners in unbound conformations. The significant improvement of the docking performance was found if reliable data on the native binding sites were available. This was possible even if data for single key amino acids at a binding interface are included. In case of binding site predictions with limited accuracy, only modest improvement compared with unbiased docking was found. The optimisation of the protocol to bias the search towards predicted binding sites was found to further improve the docking performance resulting in approximately 40% acceptable solutions within the top 10 docking predictions compared with 22% in case of unbiased docking of unbound protein structures.     

Monte Carlo replica‐exchange based ensemble docking of protein conformations

A replica‐exchange Monte Carlo (REMC) ensemble docking approach has been developed that allows efficient exploration of protein–protein docking geometries. In addition to Monte Carlo steps in translation and orientation of binding partners, possible conformational changes upon binding are included based on Monte Carlo selection of protein conformations stored as ordered pregenerated conformational ensembles. The conformational ensembles of each binding partner protein were generated by three different approaches starting from the unbound partner protein structure with a range spanning a root mean square deviation of 1–2.5 Å with respect to the unbound structure. Because MC sampling is performed to select appropriate partner conformations on the fly the approach is not limited by the number of conformations in the ensemble compared to ensemble docking of each conformer pair in ensemble cross docking. Although only a fraction of generated conformers was in closer agreement with the bound structure the REMC ensemble docking approach achieved improved docking results compared to REMC docking with only the unbound partner structures or using docking energy minimization methods. The approach has significant potential for further improvement in combination with more realistic structural ensembles and better docking scoring functions. Proteins 2017; 85:924–937. © 2016 Wiley Periodicals, Inc.     

A knowledge-based forcefield for protein-protein interface design

A distance-dependent knowledge-based potential for protein-protein interactions is derived and tested for application in protein design. Information on residue type specific C(alpha) and C(beta) pair distances is extracted from complex crystal structures in the Protein Data Bank and used in the form of radial distribution functions. The use of only backbone and C(beta) position information allows generation of relative protein-protein orientation poses with minimal sidechain information. Further coarse-graining can be done simply in the same theoretical framework to give potentials for residues of known type interacting with unknown type, as in a one-sided interface design problem. Both interface design via pose generation followed by sidechain repacking and localized protein-protein docking tests are performed on 39 nonredundant antibody-antigen complexes for which crystal structures are available. As reference, Lennard-Jones potentials, unspecific for residue type and biasing toward varying degrees of residue pair separation are used as controls. For interface design, the knowledge-based potentials give the best combination of consistently designable poses, low RMSD to the known structure, and more tightly bound interfaces with no added computational cost. 77% of the poses could be designed to give complexes with negative free energies of binding. Generally, larger interface separation promotes designability, but weakens the binding of the resulting designs. A localized docking test shows that the knowledge-based nature of the potentials improves performance and compares respectably with more sophisticated all-atoms potentials.     

A detailed comparison of current docking and scoring methods on systems of pharmaceutical relevance

A thorough evaluation of some of the most advanced docking and scoring methods currently available is described, and guidelines for the choice of an appropriate protocol for docking and virtual screening are defined. The generation of a large and highly curated test set of pharmaceutically relevant protein-ligand complexes with known binding affinities is described, and three highly regarded docking programs (Glide, GOLD, and ICM) are evaluated on the same set with respect to their ability to reproduce crystallographic binding orientations. Glide correctly identified the crystallographic pose within 2.0 A in 61% of the cases, versus 48% for GOLD and 45% for ICM. In general Glide appears to perform most consistently with respect to diversity of binding sites and ligand flexibility, while the performance of ICM and GOLD is more binding site-dependent and it is significantly poorer when binding is predominantly driven by hydrophobic interactions. The results also show that energy minimization and reranking of the top N poses can be an effective means to overcome some of the limitations of a given docking function. The same docking programs are evaluated in conjunction with three different scoring functions for their ability to discriminate actives from inactives in virtual screening. The evaluation, performed on three different systems (HIV-1 protease, IMPDH, and p38 MAP kinase), confirms that the relative performance of different docking and scoring methods is to some extent binding site-dependent. GlideScore appears to be an effective scoring function for database screening, with consistent performance across several types of binding sites, while ChemScore appears to be most useful in sterically demanding sites since it is more forgiving of repulsive interactions. Energy minimization of docked poses can significantly improve the enrichments in systems with sterically demanding binding sites. Overall Glide appears to be a safe general choice for docking, while the choice of the best scoring tool remains to a larger extent system-dependent and should be evaluated on a case-by-case basis.     

Residue conservation information for generating near-native structures in protein-protein docking

MOTIVATION: Protein-protein docking algorithms typically generate large numbers of possible complex structures with only a few of them resembling the native structure. Recently (Duan et al., Protein Sci, 14:316-218, 2005), it was observed that the surface density of conserved residue positions is high at the interface regions of interacting protein surfaces, except for antibody-antigen complexes, where a lesser number of conserved positions than average is observed at the interface regions. Using this observation, we identified putative interacting regions on the surface of interacting partners and significantly improved docking results by assigning top ranks to near-native complex structures. In this paper, we combine the residue conservation information with a widely used shape complementarity algorithm to generate candidate complex structures with a higher percentage of near-native structures (hits). What is new in this work is that the conservation information is used early in the generation stage and not only in the ranking stage of the docking algorithm. This results in a significantly larger number of generated hits and an improved predictive ability in identifying the native structure of protein-protein complexes. RESULTS: We report on results from 48 well-characterized protein complexes, which have enough residue conservation information from the same 59 benchmark complexes used in our previous work. We compute conservation indices of residue positions on the surfaces of interacting proteins using available homologous sequences from UNIPROT and calculate the solvent accessible surface area. We combine this information with shape-complementarity scores to generate candidate protein-protein complex structures. When compared with pure shape-complementarity algorithms, performed by FTDock, our method results in significantly more hits, with the improvement being over 100% in many instances. We demonstrate that residue conservation information is useful not only in refinement and scoring of docking solutions, but also helpful in enrichment of near-native-structures during the generation of candidate geometries of complex structures.     

A systematic analysis of scoring functions in rigid‐body protein docking: The delicate balance between the predictive rate improvement and the risk of overtraining

Protein‐protein interactions play fundamental roles in biological processes including signaling, metabolism, and trafficking. While the structure of a protein complex reveals crucial details about the interaction, it is often difficult to acquire this information experimentally. As the number of interactions discovered increases faster than they can be characterized, protein‐protein docking calculations may be able to reduce this disparity by providing models of the interacting proteins. Rigid‐body docking is a widely used docking approach, and is often capable of generating a pool of models within which a near‐native structure can be found. These models need to be scored in order to select the acceptable ones from the set of poses. Recently, more than 100 scoring functions from the CCharPPI server were evaluated for this task using decoy structures generated with SwarmDock. Here, we extend this analysis to identify the predictive success rates of the scoring functions on decoys from three rigid‐body docking programs, ZDOCK, FTDock, and SDOCK, allowing us to assess the transferability of the functions. We also apply set‐theoretic measure to test whether the scoring functions are capable of identifying near‐native poses within different subsets of the benchmark. This information can provide guides for the use of the most efficient scoring function for each docking method, as well as instruct future scoring functions development efforts. Proteins 2017; 85:1287–1297. © 2017 Wiley Periodicals, Inc.     

How different from random are docking predictions when ranked by scoring functions?

Docking algorithms predict the structure of protein–protein interactions. They sample the orientation of two unbound proteins to produce various predictions about their interactions, followed by a scoring step to rank the predictions. We present a statistical assessment of scoring functions used to rank near‐native orientations, applying our statistical analysis to a benchmark dataset of decoys of protein–protein complexes and assessing the statistical significance of the outcome in the Critical Assessment of PRedicted Interactions (CAPRI) scoring experiment. A P value was assigned that depended on the number of near‐native structures in the sampling. We studied the effect of filtering out redundant structures and tested the use of pair‐potentials derived using ZDock and ZRank. Our results show that for many targets, it is not possible to determine when a successful reranking performed by scoring functions results merely from random choice. This analysis reveals that changes should be made in the design of the CAPRI scoring experiment. We propose including the statistical assessment in this experiment either at the preprocessing or the evaluation step. Proteins 2010. © 2010 Wiley‐Liss, Inc.     

F2Dock: fast Fourier protein-protein docking

The functions of proteins are often realized through their mutual interactions. Determining a relative transformation for a pair of proteins and their conformations which form a stable complex, reproducible in nature, is known as docking. It is an important step in drug design, structure determination, and understanding function and structure relationships. In this paper, we extend our nonuniform fast Fourier transform-based docking algorithm to include an adaptive search phase (both translational and rotational) and thereby speed up its execution. We have also implemented a multithreaded version of the adaptive docking algorithm for even faster execution on multicore machines. We call this protein-protein docking code F2Dock (F2 = Fast Fourier). We have calibrated F2Dock based on an extensive experimental study on a list of benchmark complexes and conclude that F2Dock works very well in practice. Though all docking results reported in this paper use shape complementarity and Coulombic-potential-based scores only, F2Dock is structured to incorporate Lennard-Jones potential and reranking docking solutions based on desolvation energy .     

Score_set: A CAPRI benchmark for scoring protein complexes

Critical Assessment of PRedicted Interactions (CAPRI) has proven to be a catalyst for the development of docking algorithms. An essential step in docking is the scoring of predicted binding modes in order to identify stable complexes. In 2005, CAPRI introduced the scoring experiment, where upon completion of a prediction round, a larger set of models predicted by different groups and comprising both correct and incorrect binding modes, is made available to all participants for testing new scoring functions independently from docking calculations. Here we present an expanded benchmark data set for testing scoring functions, which comprises the consolidated ensemble of predicted complexes made available in the CAPRI scoring experiment since its inception. This consolidated scoring benchmark contains predicted complexes for 15 published CAPRI targets. These targets were subjected to 23 CAPRI assessments, due to existence of multiple binding modes for some targets. The benchmark contains more than 19,000 protein complexes. About 10% of the complexes represent docking predictions of acceptable quality or better, the remainder represent incorrect solutions (decoys). The benchmark set contains models predicted by 47 different predictor groups including web servers, which use different docking and scoring procedures, and is arguably as diverse as one may expect, representing the state of the art in protein docking. The data set is publicly available at the following URL: http://cb.iri.univ‐lille1.fr/Users/lensink/Score_set . Proteins 2014; 82:3163–3169. © 2014 Wiley Periodicals, Inc.     

An iterative knowledge-based scoring function for protein-protein recognition

Using an efficient iterative method, we have developed a distance-dependent knowledge-based scoring function to predict protein-protein interactions. The function, referred to as ITScore-PP, was derived using the crystal structures of a training set of 851 protein-protein dimeric complexes containing true biological interfaces. The key idea of the iterative method for deriving ITScore-PP is to improve the interatomic pair potentials by iteration, until the pair potentials can distinguish true binding modes from decoy modes for the protein-protein complexes in the training set. The iterative method circumvents the challenging reference state problem in deriving knowledge-based potentials. The derived scoring function was used to evaluate the ligand orientations generated by ZDOCK 2.1 and the native ligand structures on a diverse set of 91 protein-protein complexes. For the bound test cases, ITScore-PP yielded a success rate of 98.9% if the top 10 ranked orientations were considered. For the more realistic unbound test cases, the corresponding success rate was 40.7%. Furthermore, for faster orientational sampling purpose, several residue-level knowledge-based scoring functions were also derived following the similar iterative procedure. Among them, the scoring function that uses the side-chain center of mass (SCM) to represent a residue, referred to as ITScore-PP(SCM), showed the best performance and yielded success rates of 71.4% and 30.8% for the bound and unbound cases, respectively, when the top 10 orientations were considered. ITScore-PP was further tested using two other published protein-protein docking decoy sets, the ZDOCK decoy set and the RosettaDock decoy set. In addition to binding mode prediction, the binding scores predicted by ITScore-PP also correlated well with the experimentally determined binding affinities, yielding a correlation coefficient of R = 0.71 on a test set of 74 protein-protein complexes with known affinities. ITScore-PP is computationally efficient. The average run time for ITScore-PP was about 0.03 second per orientation (including optimization) on a personal computer with 3.2 GHz Pentium IV CPU and 3.0 GB RAM. The computational speed of ITScore-PP(SCM) is about an order of magnitude faster than that of ITScore-PP. ITScore-PP and/or ITScore-PP(SCM) can be combined with efficient protein docking software to study protein-protein recognition.     

Discrete molecular dynamics distinguishes nativelike binding poses from decoys in difficult targets

Virtual screening is one of the major tools used in computer-aided drug discovery. In structure-based virtual screening, the scoring function is critical to identifying the correct docking pose and accurately predicting the binding affinities of compounds. However, the performance of existing scoring functions has been shown to be uneven for different targets, and some important drug targets have proven especially challenging. In these targets, scoring functions cannot accurately identify the native or near-native binding pose of the ligand from among decoy poses, which affects both the accuracy of the binding affinity prediction and the ability of virtual screening to identify true binders in chemical libraries. Here, we present an approach to discriminating native poses from decoys in difficult targets for which several scoring functions failed to correctly identify the native pose. Our approach employs Discrete Molecular Dynamics simulations to incorporate protein-ligand dynamics and the entropic effects of binding. We analyze a collection of poses generated by docking and find that the residence time of the ligand in the native and nativelike binding poses is distinctly longer than that in decoy poses. This finding suggests that molecular simulations offer a unique approach to distinguishing the native (or nativelike) binding pose from decoy poses that cannot be distinguished using scoring functions that evaluate static structures. The success of our method emphasizes the importance of protein-ligand dynamics in the accurate determination of the binding pose, an aspect that is not addressed in typical docking and scoring protocols.     

A novel shape complementarity scoring function for protein-protein docking

Shape complementarity is the most basic ingredient of the scoring functions for protein-protein docking. Most grid-based docking algorithms use the total number of grid points at the binding interface to quantify shape complementarity. We have developed a novel Pairwise Shape Complementarity (PSC) function that is conceptually simple and rapid to compute. The favorable component of PSC is the total number of atom pairs between the receptor and the ligand within a distance cutoff. When applied to a benchmark of 49 test cases, PSC consistently ranks near-native structures higher and produces more near-native structures than the traditional grid-based function, and the improvement was seen across all prediction levels and in all categories of the benchmark. Without any post-processing or biological information about the binding site except the complementarity-determining region of antibodies, PSC predicts the complex structure correctly for 6 test cases, and ranks at least one near-native structure in the top 20 predictions for 18 test cases. Our docking program ZDOCK has been parallelized and the average computing time is 4 minutes using sixteen IBM SP3 processors. Both ZDOCK and the benchmark are freely available to academic users (http://zlab.bu.edu/~ rong/dock).     

Refinement of unbound protein docking studies using biological knowledge

In this work we present two methods for the reranking of protein-protein docking studies. One scoring method searches the InterDom database for domains that are available in the proteins to be docked and evaluates the interaction of these domains in other complexes of known structure. The second one analyzes the interface of each proposed conformation with regard to the conservation of Phe, Met, and Trp and their polar neighbor residues. The special relevance of these residues is based on a publication by Ma et al. (Proc Natl Acad Sci USA 2003;100:5772-5777), who compared the conservation of all residues in the interface region to the conservation on the rest of the protein's surface. The scoring functions were tested on 30 unbound docking test cases. The evaluation of the methods is based on the ability to rerank the output of a Fast Fourier Transformation (FFT) docking. Both were able to improve the ranking of the docking output. The best improvement was achieved for enzyme-inhibitor examples. Especially the domain-based scoring function was successful and able to place a near-native solution on one of the first six ranks for 13 of 17 (76%) enzyme-inhibitor complexes [in 53% (nine complexes) even on the first rank]. The method evaluating residue conservation allowed us to increase the number of good solutions within the first 100 ranks out of approximately 9000 in 82% of the 17 enzyme-inhibitor test cases, and for seven (41%) out of 17 enzyme-inhibitor complexes, a near native solution was placed within the first seven ranks.