Novel sampling strategies and a coarse-grained score function for docking homomers,flexible heteromers,and oligosaccharides using Rosetta in CAPRI rounds 37–45

Critical Assessment of PRediction of Interactions (CAPRI) rounds 37 through 45 introduced larger complexes, new macromolecules, and multistage assemblies. For these rounds, we used and expanded docking methods in Rosetta to model 23 target complexes. We successfully predicted 14 target complexes and recognized and refined near-native models generated by other groups for two further targets. Notably, for targets T110 and T136, we achieved the closest prediction of any CAPRI participant. We created several innovative approaches during these rounds. Since round 39 (target 122), we have used the new RosettaDock 4.0, which has a revamped coarse-grained energy function and the ability to perform conformer selection during docking with hundreds of pregenerated protein backbones. Ten of the complexes had some degree of symmetry in their interactions, so we tested Rosetta SymDock, realized its shortcomings, and developed the next-generation symmetric docking protocol, SymDock2, which includes docking of multiple backbones and induced-fit refinement. Since the last CAPRI assessment, we also developed methods for modeling and designing carbohydrates in Rosetta, and we used them to successfully model oligosaccharide-protein complexes in round 41. Although the results were broadly encouraging, they also highlighted the pressing need to invest in (a) flexible docking algorithms with the ability to model loop and linker motions and in (b) new sampling and scoring methods for oligosaccharide-protein interactions.     

Automatic prediction of flexible regions improves the accuracy of protein-protein docking models

Computational models of protein-protein docking that incorporate backbone flexibility can predict perturbations of the backbone and side chains during docking and produce protein interaction models with atomic accuracy. Most previous models usually predefine flexible regions by visually comparing the bound and unbound structures. In this paper, we propose a general method to automatically identify the flexible hinges for domain assembly and the flexible loops for loop refinement, in addition to predicting the corresponding movements of the identified active residues. We conduct experiments to evaluate performance of our approach on two test sets. Comparison of results on test set I between algorithms with and without prediction of flexible regions demonstrate the superior recovery of energy funnels in many target interactions using the new loop refinement model. In addition, our decoys are superior for each target. Indeed, the total number of satisfactory models is almost double that of other programs. The results on test set II docking tests produced by our domain assembly method also show encouraging results. Of the three targets examined, one exhibits energy funnel and the best models of the other two targets all meet the conditions of acceptable accuracy. Results demonstrate that the automatic prediction of flexible backbone regions can greatly improve the performance of protein-protein docking models.     

Performance of human and server prediction in CAPRI rounds 38-45

CAPRI challenges offer a variety of blind tests for protein-protein interaction prediction. In CAPRI Rounds 38-45, we generated a set of putative binding modes for each target with an FFT-based docking algorithm, and then scored and ranked these binding modes with a proprietary scoring function, ITScorePP. We have also developed a novel web server, Rebipp. The algorithm utilizes information retrieval to identify relevant biological information to significantly reduce the search space for a particular protein. In parallel, we have also constructed a GPU-based docking server, MDockPP, for protein-protein complex structure prediction. Here, the performance of our protocol in CAPRI rounds 38-45 is reported, which include 16 docking and scoring targets. Among them, three targets contain multiple interfaces: Targets 124, 125, and 136 have 2, 4, and 3 interfaces, respectively. In the predictor experiments, we predicted correct binding modes for nine targets, including one high-accuracy interface, six medium-accuracy binding modes, and six acceptable-accuracy binding modes. For the docking server prediction experiments, we predicted correct binding modes for eight targets, including one high-accuracy, three medium-accuracy, and five acceptable-accuracy binding modes.     

Performance of ZDOCK and ZRANK in CAPRI rounds 13-19

We report the performance of the ZDOCK and ZRANK algorithms in CAPRI rounds 13-19 and introduce a novel measure atom contact frequency (ACF). To compute ACF, we identify the residues that most often make contact with the binding partner in the complete set of ZDOCK predictions for each target. We used ACF to predict the interface of the proteins, which, in combination with the biological data available in the literature, is a valuable addition to our docking pipeline. Furthermore, we incorporated a straightforward and efficient clustering algorithm with two purposes: (1) to determine clusters of similar docking poses (corresponding to energy funnels) and (2) to remove redundancies from the final set of predictions. With these new developments, we achieved at least one acceptable prediction for targets 29 and 36, at least one medium-quality prediction for targets 41 and 42, and at least one high-quality prediction for targets 37 and 40; thus, we succeeded for six out of a total of 12 targets.     

Scoring docking models with evolutionary information

We have developed methods for the extraction of evolutionary information from multiple sequence alignments for use in the study of the evolution of protein interaction networks and in the prediction of protein interaction. For Rounds 3, 4, and 5 of the CAPRI experiment, we used scores derived from the analysis of multiple sequence alignments to submit predictions for 7 of the 12 targets. Our docking models were generated with Hex and GRAMM, but all our predictions were selected using methods based on multiple sequence alignments and on the available experimental evidence. With this approach, we were able to predict acceptable level models for 4 of the targets, and for a fifth target, we located the residues involved in the binding surface. Here we detail our successes and highlight several of the limitations and problems that we faced while dealing with particular docking cases.     

Benchmarking and analysis of protein docking performance in Rosetta v3.2

RosettaDock has been increasingly used in protein docking and design strategies in order to predict the structure of protein-protein interfaces. Here we test capabilities of RosettaDock 3.2, part of the newly developed Rosetta v3.2 modeling suite, against Docking Benchmark 3.0, and compare it with RosettaDock v2.3, the latest version of the previous Rosetta software package. The benchmark contains a diverse set of 116 docking targets including 22 antibody-antigen complexes, 33 enzyme-inhibitor complexes, and 60 'other' complexes. These targets were further classified by expected docking difficulty into 84 rigid-body targets, 17 medium targets, and 14 difficult targets. We carried out local docking perturbations for each target, using the unbound structures when available, in both RosettaDock v2.3 and v3.2. Overall the performances of RosettaDock v2.3 and v3.2 were similar. RosettaDock v3.2 achieved 56 docking funnels, compared to 49 in v2.3. A breakdown of docking performance by protein complex type shows that RosettaDock v3.2 achieved docking funnels for 63% of antibody-antigen targets, 62% of enzyme-inhibitor targets, and 35% of 'other' targets. In terms of docking difficulty, RosettaDock v3.2 achieved funnels for 58% of rigid-body targets, 30% of medium targets, and 14% of difficult targets. For targets that failed, we carry out additional analyses to identify the cause of failure, which showed that binding-induced backbone conformation changes account for a majority of failures. We also present a bootstrap statistical analysis that quantifies the reliability of the stochastic docking results. Finally, we demonstrate the additional functionality available in RosettaDock v3.2 by incorporating small-molecules and non-protein co-factors in docking of a smaller target set. This study marks the most extensive benchmarking of the RosettaDock module to date and establishes a baseline for future research in protein interface modeling and structure prediction.     

Evaluation of AlphaFold2 structures as docking targets

AlphaFold2 is a promising new tool for researchers to predict protein structures and generate high-quality models, with low backbone and global root-mean-square deviation (RMSD) when compared with experimental structures. However, it is unclear if the structures predicted by AlphaFold2 will be valuable targets of docking. To address this question, we redocked ligands in the PDBbind datasets against the experimental co-crystallized receptor structures and against the AlphaFold2 structures using AutoDock-GPU. We find that the quality measure provided during structure prediction is not a good predictor of docking performance, despite accurately reflecting the quality of the alpha carbon alignment with experimental structures. Removing low-confidence regions of the predicted structure and making side chains flexible improves the docking outcomes. Overall, despite high-quality prediction of backbone conformation, fine structural details limit the naive application of AlphaFold2 models as docking targets.     

Incorporating biochemical information and backbone flexibility in RosettaDock for CAPRI rounds 6-12

In CAPRI rounds 6-12, RosettaDock successfully predicted 2 of 5 unbound-unbound targets to medium accuracy. Improvement over the previous method was achieved with computational mutagenesis to select decoys that match the energetics of experimentally determined hot spots. In the case of Target 21, Orc1/Sir1, this resulted in a successful docking prediction where RosettaDock alone or with simple site constraints failed. Experimental information also helped limit the interacting region of TolB/Pal, producing a successful prediction of Target 26. In addition, we docked multiple loop conformations for Target 20, and we developed a novel flexible docking algorithm to simultaneously optimize backbone conformation and rigid-body orientation to generate a wide diversity of conformations for Target 24. Continued challenges included docking of homology targets that differ substantially from their template (sequence identity <50%) and accounting for large conformational changes upon binding. Despite a larger number of unbound-unbound and homology model binding targets, Rounds 6-12 reinforced that RosettaDock is a powerful algorithm for predicting bound complex structures, especially when combined with experimental data.     

Successful discrimination of protein interactions

We present results from the prediction of protein complexes associated with the first Critical Assessment of PRediction of Interactions (CAPRI) experiment. Our algorithm, SmoothDock, comprises four steps: (1) we perform rigid body docking using the program DOT, keeping the top 20,000 structures as ranked by surface complementarity; (2) we rerank these structures according to a free energy estimate that includes both desolvation and electrostatics and retain the top 2000 complexes; (3) we cluster the filtered complexes using a pairwise root-mean-square deviation (RMSD) criterion; (4) the 25 largest clusters are subject to a smooth docking discrimination algorithm where van der Waals forces are taken into account. We predicted targets 1, 6, and 7 with RMSDs of 9.5, 2.4, and 2.6 A, respectively. More importantly, from the perspective of biological applications, our approach consistently ranked the correct model first (i.e., with highest confidence). For target 5 we identified the binding region but not the correct orientation. Although we were able to find reasonable clusters for all targets, low-affinity complexes (K(d) < nM) were harder to discriminate. For four of seven targets, the top models predicted by our automated procedure were among the best communitywide predictions.     

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.     

Docking and scoring with alternative side-chain conformations

We describe a scoring and modeling procedure for docking ligands into protein models that have either modeled or flexible side-chain conformations. Our methodical contribution comprises a procedure for generating new potentials of mean force for the ROTA scoring function which we have introduced previously for optimizing side-chain conformations with the tool IRECS. The ROTA potentials are specially trained to tolerate small-scale positional errors of atoms that are characteristic of (i) side-chain conformations that are modeled using a sparse rotamer library and (ii) ligand conformations that are generated using a docking program. We generated both rigid and flexible protein models with our side-chain prediction tool IRECS and docked ligands to proteins using the scoring function ROTA and the docking programs FlexX (for rigid side chains) and FlexE (for flexible side chains). We validated our approach on the forty screening targets of the DUD database. The validation shows that the ROTA potentials are especially well suited for estimating the binding affinity of ligands to proteins. The results also show that our procedure can compensate for the performance decrease in screening that occurs when using protein models with side chains modeled with a rotamer library instead of using X-ray structures. The average runtime per ligand of our method is 168 seconds on an Opteron V20z, which is fast enough to allow virtual screening of compound libraries for drug candidates.     

Geometry-based flexible and symmetric protein docking

We present a set of geometric docking algorithms for rigid, flexible, and cyclic symmetry docking. The algorithms are highly efficient and have demonstrated very good performance in CAPRI Rounds 3-5. The flexible docking algorithm, FlexDock, is unique in its ability to handle any number of hinges in the flexible molecule, without degradation in run-time performance, as compared to rigid docking. The algorithm for reconstruction of cyclically symmetric complexes successfully assembles multimolecular complexes satisfying C(n) symmetry for any n in a matter of minutes on a desktop PC. Most of the algorithms presented here are available at the Tel Aviv University Structural Bioinformatics Web server (http://bioinfo3d.cs.tau.ac.il/).     

Docking prediction using biological information, ZDOCK sampling technique, and clustering guided by the DFIRE statistical energy function

We entered the CAPRI experiment during the middle of Round 4 and have submitted predictions for all 6 targets released since then. We used the following procedures for docking prediction: (1) the identification of possible binding region(s) of a target based on known biological information, (2) rigid-body sampling around the binding region(s) by using the docking program ZDOCK, (3) ranking of the sampled complex conformations by employing the DFIRE-based statistical energy function, (4) clustering based on pairwise root-mean-square distance and the DFIRE energy, and (5) manual inspection and relaxation of the side-chain conformations of the top-ranked structures by geometric constraint. Reasonable predictions were made for 4 of the 6 targets. The best fraction of native contacts within the top 10 models are 89.1% for Target 12, 54.3% for Target 13, 29.3% for Target 14, and 94.1% for Target 18. The origin of successes and failures is discussed. .     

Performance and enhancement of the LZerD protein assembly pipeline in CAPRI 38-46

We report the performance of the protein docking prediction pipeline of our group and the results for Critical Assessment of Prediction of Interactions (CAPRI) rounds 38-46. The pipeline integrates programs developed in our group as well as other existing scoring functions. The core of the pipeline is the LZerD protein-protein docking algorithm. If templates of the target complex are not found in PDB, the first step of our docking prediction pipeline is to run LZerD for a query protein pair. Meanwhile, in the case of human group prediction, we survey the literature to find information that can guide the modeling, such as protein-protein interface information. In addition to any literature information and binding residue prediction, generated docking decoys were selected by a rank aggregation of statistical scoring functions. The top 10 decoys were relaxed by a short molecular dynamics simulation before submission to remove atom clashes and improve side-chain conformations. In these CAPRI rounds, our group, particularly the LZerD server, showed robust performance. On the other hand, there are failed cases where some other groups were successful. To understand weaknesses of our pipeline, we analyzed sources of errors for failed targets. Since we noted that structure refinement is a step that needs improvement, we newly performed a comparative study of several refinement approaches. Finally, we show several examples that illustrate successful and unsuccessful cases by our group.     

Accelerating protein docking in ZDOCK using an advanced 3D convolution library

Computational prediction of the 3D structures of molecular interactions is a challenging area, often requiring significant computational resources to produce structural predictions with atomic-level accuracy. This can be particularly burdensome when modeling large sets of interactions, macromolecular assemblies, or interactions between flexible proteins. We previously developed a protein docking program, ZDOCK, which uses a fast Fourier transform to perform a 3D search of the spatial degrees of freedom between two molecules. By utilizing a pairwise statistical potential in the ZDOCK scoring function, there were notable gains in docking accuracy over previous versions, but this improvement in accuracy came at a substantial computational cost. In this study, we incorporated a recently developed 3D convolution library into ZDOCK, and additionally modified ZDOCK to dynamically orient the input proteins for more efficient convolution. These modifications resulted in an average of over 8.5-fold improvement in running time when tested on 176 cases in a newly released protein docking benchmark, as well as substantially less memory usage, with no loss in docking accuracy. We also applied these improvements to a previous version of ZDOCK that uses a simpler non-pairwise atomic potential, yielding an average speed improvement of over 5-fold on the docking benchmark, while maintaining predictive success. This permits the utilization of ZDOCK for more intensive tasks such as docking flexible molecules and modeling of interactomes, and can be run more readily by those with limited computational resources.     

The interprotein scoring noises in glide docking scores

Small molecule drugs are rarely selective enough to interact solely with their designated targets. Unintended "off-target" interactions often lead to side effects, but also serendipitously lead to new therapeutic uses. Identification of the off-targets of a compound is therefore of significant value to the evaluation of its developmental potential. In computational biology, the strategy of "reverse docking" has been introduced to predict the targets of a compound, which uses a compound to virtually screen a library of proteins, reversing the bait and prey in "normal" docking screenings. The present study shows that, in reverse docking, additional optimization of the scoring function may help to improve the target prediction accuracy. In a case study with the Glide scores, we found that only 57% of the ligand-protein relationships could be correctly identified in a library of 58 complexes whose crystal binding conformations were all able to be accurately reproduced. This was likely a result of the constant over- or under-estimation of the scores for specific proteins. In other words, there were interprotein noises in the Glide scores. Introducing a correction term based on protein characteristics improved the target-prediction accuracy by 27% (57-72%). It is our hope that this focused discussion on the Glide scores would invite further efforts to characterize and normalize this type of interprotein noises in all docking scores, so that better target prediction accuracy can be achieved with the strategy of reverse docking.     

SymmRef: a flexible refinement method for symmetric multimers

Symmetric protein complexes are abundant in the living cell. Predicting their atomic structure can shed light on the mechanism of many important biological processes. Symmetric docking methods aim to predict the structure of these complexes given the unbound structure of a single monomer, or its model. Symmetry constraints reduce the search-space of these methods and make the prediction easier compared to asymmetric protein-protein docking. However, the challenge of modeling the conformational changes that the monomer might undergo is a major obstacle. In this article, we present SymmRef, a novel method for refinement and reranking of symmetric docking solutions. The method models backbone and side-chain movements and optimizes the rigid-body orientations of the monomers. The backbone movements are modeled by normal modes minimization and the conformations of the side-chains are modeled by selecting optimal rotamers. Since solved structures of symmetric multimers show asymmetric side-chain conformations, we do not use symmetry constraints in the side-chain optimization procedure. The refined models are re-ranked according to an energy score. We tested the method on a benchmark of unbound docking challenges. The results show that the method significantly improves the accuracy and the ranking of symmetric rigid docking solutions. SymmRef is available for download at http:// bioinfo3d.cs.tau.ac.il/SymmRef/download.html.     

Prediction of the unknown: inspiring experience with the CAPRI experiment

We submitted predictions for all seven targets in the CAPRI experiment. For four targets, our submitted models included acceptable, medium accuracy predictions of the structures of the complexes, and for a fifth target we identified the location of the binding site of one of the molecules. We used a weighted-geometric docking algorithm in which contacts involving specified parts of the surfaces of either one or both molecules were up-weighted or down-weighted. The weights were based on available structural and biochemical data or on sequence analyses. The weighted-geometric docking proved very useful for five targets, improving the complementarity scores and the ranks of the nearly correct solutions, as well as their statistical significance. In addition, the weighted-geometric docking promoted formation of clusters of similar solutions, which include more accurate predictions.