Energy-based graph convolutional networks for scoring protein docking models

Structural information about protein-protein interactions, often missing at the interactome scale, is important for mechanistic understanding of cells and rational discovery of therapeutics. Protein docking provides a computational alternative for such information. However, ranking near-native docked models high among a large number of candidates, often known as the scoring problem, remains a critical challenge. Moreover, estimating model quality, also known as the quality assessment problem, is rarely addressed in protein docking. In this study, the two challenging problems in protein docking are regarded as relative and absolute scoring, respectively, and addressed in one physics-inspired deep learning framework. We represent protein and complex structures as intra- and inter-molecular residue contact graphs with atom-resolution node and edge features. And we propose a novel graph convolutional kernel that aggregates interacting nodes’ features through edges so that generalized interaction energies can be learned directly from 3D data. The resulting energy-based graph convolutional networks (EGCN) with multihead attention are trained to predict intra- and inter-molecular energies, binding affinities, and quality measures (interface RMSD) for encounter complexes. Compared to a state-of-the-art scoring function for model ranking, EGCN significantly improves ranking for a critical assessment of predicted interactions (CAPRI) test set involving homology docking; and is comparable or slightly better for Score_set, a CAPRI benchmark set generated by diverse community-wide docking protocols not known to training data. For Score_set quality assessment, EGCN shows about 27% improvement to our previous efforts. Directly learning from 3D structure data in graph representation, EGCN represents the first successful development of graph convolutional networks for protein docking.     

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 with backbone flexibility

Computational protein-protein docking methods currently can create models with atomic accuracy for protein complexes provided that the conformational changes upon association are restricted to the side chains. However, it remains very challenging to account for backbone conformational changes during docking, and most current methods inherently keep monomer backbones rigid for algorithmic simplicity and computational efficiency. Here we present a reformulation of the Rosetta docking method that incorporates explicit backbone flexibility in protein-protein docking. The new method is based on a "fold-tree" representation of the molecular system, which seamlessly integrates internal torsional degrees of freedom and rigid-body degrees of freedom. Problems with internal flexible regions ranging from one or more loops or hinge regions to all of one or both partners can be readily treated using appropriately constructed fold trees. The explicit treatment of backbone flexibility improves both sampling in the vicinity of the native docked conformation and the energetic discrimination between near-native and incorrect models.     

Electrostatic contributions to protein-protein interactions: fast energetic filters for docking and their physical basis

The methods of continuum electrostatics are used to calculate the binding free energies of a set of protein-protein complexes including experimentally determined structures as well as other orientations generated by a fast docking algorithm. In the native structures, charged groups that are deeply buried were often found to favor complex formation (relative to isosteric nonpolar groups), whereas in nonnative complexes generated by a geometric docking algorithm, they were equally likely to be stabilizing as destabilizing. These observations were used to design a new filter for screening docked conformations that was applied, in conjunction with a number of geometric filters that assess shape complementarity, to 15 antibody-antigen complexes and 14 enzyme-inhibitor complexes. For the bound docking problem, which is the major focus of this paper, native and near-native solutions were ranked first or second in all but two enzyme-inhibitor complexes. Less success was encountered for antibody-antigen complexes, but in all cases studied, the more complete free energy evaluation was able to identify native and near-native structures. A filter based on the enrichment of tyrosines and tryptophans in antibody binding sites was applied to the antibody-antigen complexes and resulted in a native and near-native solution being ranked first and second in all cases. A clear improvement over previously reported results was obtained for the unbound antibody-antigen examples as well. The algorithm and various filters used in this work are quite efficient and are able to reduce the number of plausible docking orientations to a size small enough so that a final more complete free energy evaluation on the reduced set becomes computationally feasible.     

H‐bond network optimization in protein–protein complexes: Are all‐atom force field scores enough?

Structural prediction of protein-protein complexes given the structures of the two interacting compounds in their unbound state is a key problem in biophysics. In addition to the problem of sampling of near-native orientations, one of the modeling main difficulties is to discriminate true from false positives. Here, we present a hierarchical protocol for docking refinement able to discriminate near native poses from a group of docking candidates. The main idea is to combine an efficient sampling of the full system hydrogen bond network and side chains, together with an all-atom force field and a surface generalized born implicit solvent. We tested our method on a set of twenty two complexes containing a near-native solution within the top 100 docking poses, obtaining a near native solution as the top pose in 70% of the cases. We show that all atom force fields optimized H-bond networks do improve significantly state of the art scoring functions.     

Protein docking by Rotation-Based Uniform Sampling (RotBUS) with fast computing of intermolecular contact distance and residue desolvation

Background  

Protein-protein interactions are fundamental for the majority of cellular processes and their study is of enormous biotechnological and therapeutic interest. In recent years, a variety of computational approaches to the protein-protein docking problem have been reported, with encouraging results. Most of the currently available protein-protein docking algorithms are composed of two clearly defined parts: the sampling of the rotational and translational space of the interacting molecules, and the scoring and clustering of the resulting orientations. Although this kind of strategy has shown some of the most successful results in the CAPRI blind test , more efforts need to be applied. Thus, the sampling protocol should generate a pool of conformations that include a sufficient number of near-native ones, while the scoring function should discriminate between near-native and non-near-native proposed conformations. On the other hand, protocols to efficiently include full flexibility on the protein structures are increasingly needed.     

pyDock: electrostatics and desolvation for effective scoring of rigid-body protein-protein docking

The accurate scoring of rigid-body docking orientations represents one of the major difficulties in protein-protein docking prediction. Other challenges are the development of faster and more efficient sampling methods and the introduction of receptor and ligand flexibility during simulations. Overall, good discrimination of near-native docking poses from the very early stages of rigid-body protein docking is essential step before applying more costly interface refinement to the correct docking solutions. Here we explore a simple approach to scoring of rigid-body docking poses, which has been implemented in a program called pyDock. The scheme is based on Coulombic electrostatics with distance dependent dielectric constant, and implicit desolvation energy with atomic solvation parameters previously adjusted for rigid-body protein-protein docking. This scoring function is not highly dependent on specific geometry of the docking poses and therefore can be used in rigid-body docking sets generated by a variety of methods. We have tested the procedure in a large benchmark set of 80 unbound docking cases. The method is able to detect a near-native solution from 12,000 docking poses and place it within the 100 lowest-energy docking solutions in 56% of the cases, in a completely unrestricted manner and without any other additional information. More specifically, a near-native solution will lie within the top 20 solutions in 37% of the cases. The simplicity of the approach allows for a better understanding of the physical principles behind protein-protein association, and provides a fast tool for the evaluation of large sets of rigid-body docking poses in search of the near-native orientation.     

Optimal docking area: a new method for predicting protein-protein interaction sites

Understanding energetics and mechanism of protein-protein association remains one of the biggest theoretical problems in structural biology. It is assumed that desolvation must play an essential role during the association process, and indeed protein-protein interfaces in obligate complexes have been found to be highly hydrophobic. However, the identification of protein interaction sites from surface analysis of proteins involved in non-obligate protein-protein complexes is more challenging. Here we present Optimal Docking Area (ODA), a new fast and accurate method of analyzing a protein surface in search of areas with favorable energy change when buried upon protein-protein association. The method identifies continuous surface patches with optimal docking desolvation energy based on atomic solvation parameters adjusted for protein-protein docking. The procedure has been validated on the unbound structures of a total of 66 non-homologous proteins involved in non-obligate protein-protein hetero-complexes of known structure. Optimal docking areas with significant low-docking surface energy were found in around half of the proteins. The 'ODA hot spots' detected in X-ray unbound structures were correctly located in the known protein-protein binding sites in 80% of the cases. The role of these low-surface-energy areas during complex formation is discussed. Burial of these regions during protein-protein association may favor the complexed configurations with near-native interfaces but otherwise arbitrary orientations, thus driving the formation of an encounter complex. The patch prediction procedure is freely accessible at http://www.molsoft.com/oda and can be easily scaled up for predictions in structural proteomics.     

Protein-Protein Docking: From Interaction to Interactome

The protein-protein docking problem is one of the focal points of activity in computational biophysics and structural biology. The three-dimensional structure of a protein-protein complex, generally, is more difficult to determine experimentally than the structure of an individual protein. Adequate computational techniques to model protein interactions are important because of the growing number of known protein structures, particularly in the context of structural genomics. Docking offers tools for fundamental studies of protein interactions and provides a structural basis for drug design. Protein-protein docking is the prediction of the structure of the complex, given the structures of the individual proteins. In the heart of the docking methodology is the notion of steric and physicochemical complementarity at the protein-protein interface. Originally, mostly high-resolution, experimentally determined (primarily by x-ray crystallography) protein structures were considered for docking. However, more recently, the focus has been shifting toward lower-resolution modeled structures. Docking approaches have to deal with the conformational changes between unbound and bound structures, as well as the inaccuracies of the interacting modeled structures, often in a high-throughput mode needed for modeling of large networks of protein interactions. The growing number of docking developers is engaged in the community-wide assessments of predictive methodologies. The development of more powerful and adequate docking approaches is facilitated by rapidly expanding information and data resources, growing computational capabilities, and a deeper understanding of the fundamental principles of protein interactions.     

Protein-Protein Docking: From Interaction to Interactome

The protein-protein docking problem is one of the focal points of activity in computational biophysics and structural biology. The three-dimensional structure of a protein-protein complex, generally, is more difficult to determine experimentally than the structure of an individual protein. Adequate computational techniques to model protein interactions are important because of the growing number of known protein structures, particularly in the context of structural genomics. Docking offers tools for fundamental studies of protein interactions and provides a structural basis for drug design. Protein-protein docking is the prediction of the structure of the complex, given the structures of the individual proteins. In the heart of the docking methodology is the notion of steric and physicochemical complementarity at the protein-protein interface. Originally, mostly high-resolution, experimentally determined (primarily by x-ray crystallography) protein structures were considered for docking. However, more recently, the focus has been shifting toward lower-resolution modeled structures. Docking approaches have to deal with the conformational changes between unbound and bound structures, as well as the inaccuracies of the interacting modeled structures, often in a high-throughput mode needed for modeling of large networks of protein interactions. The growing number of docking developers is engaged in the community-wide assessments of predictive methodologies. The development of more powerful and adequate docking approaches is facilitated by rapidly expanding information and data resources, growing computational capabilities, and a deeper understanding of the fundamental principles of protein interactions.     

Consensus scoring for enriching near-native structures from protein-protein docking decoys

The identification of near native protein-protein complexes among a set of decoys remains highly challenging. A strategy for improving the success rate of near native detection is to enrich near native docking decoys in a small number of top ranked decoys. Recently, we found that a combination of three scoring functions (energy, conservation, and interface propensity) can predict the location of binding interface regions with reasonable accuracy. Here, these three scoring functions are modified and combined into a consensus scoring function called ENDES for enriching near native docking decoys. We found that all individual scores result in enrichment for the majority of 28 targets in ZDOCK2.3 decoy set and the 22 targets in Benchmark 2.0. Among the three scores, the interface propensity score yields the highest enrichment in both sets of protein complexes. When these scores are combined into the ENDES consensus score, a significant increase in enrichment of near-native structures is found. For example, when 2000 dock decoys are reduced to 200 decoys by ENDES, the fraction of near-native structures in docking decoys increases by a factor of about six in average. ENDES was implemented into a computer program that is available for download at http://sparks.informatics.iupui.edu.     

Structural similarity and classification of protein interaction interfaces

Interactions between proteins play a key role in many cellular processes. Studying protein-protein interactions that share similar interaction interfaces may shed light on their evolution and could be helpful in elucidating the mechanisms behind stability and dynamics of the protein complexes. When two complexes share structurally similar subunits, the similarity of the interaction interfaces can be found through a structural superposition of the subunits. However, an accurate detection of similarity between the protein complexes containing subunits of unrelated structure remains an open problem. Here, we present an alignment-free machine learning approach to measure interface similarity. The approach relies on the feature-based representation of protein interfaces and does not depend on the superposition of the interacting subunit pairs. Specifically, we develop an SVM classifier of similar and dissimilar interfaces and derive a feature-based interface similarity measure. Next, the similarity measure is applied to a set of 2,806×2,806 binary complex pairs to build a hierarchical classification of protein-protein interactions. Finally, we explore case studies of similar interfaces from each level of the hierarchy, considering cases when the subunits forming interactions are either homologous or structurally unrelated. The analysis has suggested that the positions of charged residues in the homologous interfaces are not necessarily conserved and may exhibit more complex conservation patterns.     

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.     

A Unified Conformational Selection and Induced Fit Approach to Protein-Peptide Docking

Protein-peptide interactions are vital for the cell. They mediate, inhibit or serve as structural components in nearly 40% of all macromolecular interactions, and are often associated with diseases, making them interesting leads for protein drug design. In recent years, large-scale technologies have enabled exhaustive studies on the peptide recognition preferences for a number of peptide-binding domain families. Yet, the paucity of data regarding their molecular binding mechanisms together with their inherent flexibility makes the structural prediction of protein-peptide interactions very challenging. This leaves flexible docking as one of the few amenable computational techniques to model these complexes. We present here an ensemble, flexible protein-peptide docking protocol that combines conformational selection and induced fit mechanisms. Starting from an ensemble of three peptide conformations (extended, a-helix, polyproline-II), flexible docking with HADDOCK generates 79.4% of high quality models for bound/unbound and 69.4% for unbound/unbound docking when tested against the largest protein-peptide complexes benchmark dataset available to date. Conformational selection at the rigid-body docking stage successfully recovers the most relevant conformation for a given protein-peptide complex and the subsequent flexible refinement further improves the interface by up to 4.5 Å interface RMSD. Cluster-based scoring of the models results in a selection of near-native solutions in the top three for ∼75% of the successfully predicted cases. This unified conformational selection and induced fit approach to protein-peptide docking should open the route to the modeling of challenging systems such as disorder-order transitions taking place upon binding, significantly expanding the applicability limit of biomolecular interaction modeling by docking.     

The impact of protein flexibility on protein-protein docking

Accounting for protein flexibility in protein-protein docking algorithms is challenging, and most algorithms therefore treat proteins as rigid bodies or permit side-chain motion only. While the consequences are obvious when there are large conformational changes upon binding, the situation is less clear for the modest conformational changes that occur upon formation of most protein-protein complexes. We have therefore studied the impact of local protein flexibility on protein-protein association by means of rigid body and torsion angle dynamics simulation. The binding of barnase and barstar was chosen as a model system for this study, because the complexation of these 2 proteins is well-characterized experimentally, and the conformational changes accompanying binding are modest. On the side-chain level, we show that the orientation of particular residues at the interface (so-called hotspot residues) have a crucial influence on the way contacts are established during docking from short protein separations of approximately 5 A. However, side-chain torsion angle dynamics simulations did not result in satisfactory docking of the proteins when using the unbound protein structures. This can be explained by our observations that, on the backbone level, even small (2 A) local loop deformations affect the dynamics of contact formation upon docking. Complementary shape-based docking calculations confirm this result, which indicates that both side-chain and backbone levels of flexibility influence short-range protein-protein association and should be treated simultaneously for atomic-detail computational docking of proteins.     

Re-Docking Scheme for Generating Near-Native Protein Complexes by Assembling Residue Interaction Fingerprints

Interaction profile method is a useful method for processing rigid-body docking. After the docking process, the resulting set of docking poses could be classified by calculating similarities among them using these interaction profiles to search for near-native poses. However, there are some cases where the near-native poses are not included in this set of docking poses even when the bound-state structures are used. Therefore, we have developed a method for generating near-native docking poses by introducing a re-docking process. We devised a method for calculating the profile of interaction fingerprints by assembling protein complexes after determining certain core-protein complexes. For our analysis, we used 44 bound-state protein complexes selected from the ZDOCK benchmark dataset ver. 2.0, including some protein pairs none of which generated near-native poses in the docking process. Consequently, after the re-docking process we obtained profiles of interaction fingerprints, some of which yielded near-native poses. The re-docking process involved searching for possible docking poses in a restricted area using the profile of interaction fingerprints. If the profile includes interactions identical to those in the native complex, we obtained near-native docking poses. Accordingly, near-native poses were obtained for all bound-state protein complexes examined here. Application of interaction fingerprints to the re-docking process yielded structures with more native interactions, even when a docking pose, obtained following the initial docking process, contained only a small number of native amino acid interactions. Thus, utilization of the profile of interaction fingerprints in the re-docking process yielded more near-native poses.     

基于支持向量机的蛋白质相互作用识别

支持向量机（SVM）是广泛应用于各个领域的分类算法,包括生物信息学。本研究应用SVM作为蛋白质相互作用的分类算法,所用蛋白质相互作用数据下载于墨尼黑生物信息学中心的酿酒酵母数据集,包含有6736条蛋白质,其中相互作用的有4837对,不相互作用的有9674对。提取蛋白质主要结构的电荷和等电位点特征,并应用SVM分类算法对此进行了分类。结果显示,分类的正确率在60％左右,但是较系统发育谱法还是获得了较高的分类正确率。     

Optimal design of protein docking potentials: efficiency and limitations

Protein-protein docking is a challenging computational problem in functional genomics, particularly when one or both proteins undergo conformational change(s) upon binding. The major challenge is to define scoring function soft enough to tolerate these changes and specific enough to distinguish between near-native and "misdocked" conformations. Using a linear programming technique, we derived protein docking potentials (PDPs) that comply with this requirement. We considered a set of 63 nonredundant complexes to this aim, and generated 400,000 putative docked complexes (decoys) based on shape complementarity criterion for each complex. The PDPs were required to yield for the native (correctly docked) structure a potential energy lower than those of all the nonnative (misdocked) structures. The energy constraints applied to all complexes led to ca. 25 million inequalities, the simultaneous solution of which yielded an optimal set of PDPs that discriminated the correctly docked (up to 4.0 A root-mean-square deviation from known complex structure) structure among the 85 top-ranking (0.02%) decoys in 59/63 examined bound-bound cases. The high performance of the potentials was further verified in jackknife tests and by ranking putative docked conformation submitted to CAPRI. In addition to their utility in identifying correctly folded complexes, the PDPs reveal biologically meaningful features that distinguish docking potentials from folding potentials.