Multi-LZerD: multiple protein docking for asymmetric complexes

Integrative modeling meets deep learning: Recent advances in modeling protein assemblies

Recent progress in protein structure prediction based on deep learning revolutionized the field of Structural Biology. Beyond single proteins, it also enabled high-throughput prediction of structures of protein-protein interactions. Despite the success in predicting complex structures, large macromolecular assemblies still require specialized approaches. Here we describe recent advances in modeling macromolecular assemblies using integrative and hierarchical approaches. We highlight applications that predict protein-protein interactions and challenges in modeling complexes based on the interaction networks, including the prediction of complex stoichiometry and heterogeneity.

Review and Comparative Analysis of Methods and Advancements in Predicting Protein Complex Structure

Protein complexes perform diverse biological functions, and obtaining their three-dimensional structure is critical to understanding and grasping their functions. In many cases, it's not just two proteins interacting to form a dimer; instead, multiple proteins interact to form a multimer. Experimentally resolving protein complex structures can be quite challenging. Recently, there have been efforts and methods that build upon prior predictions of dimer structures to attempt to predict multimer structures. However, in comparison to monomeric protein structure prediction, the accuracy of protein complex structure prediction remains relatively low. This paper provides an overview of recent advancements in efficient computational models for predicting protein complex structures. We introduce protein-protein docking methods in detail and summarize their main ideas, applicable modes, and related information. To enhance prediction accuracy, other critical protein-related information is also integrated, such as predicting interchain residue contact, utilizing experimental data like cryo-EM experiments, and considering protein interactions and non-interactions. In addition, we comprehensively review computational approaches for end-to-end prediction of protein complex structures based on artificial intelligence (AI) technology and describe commonly used datasets and representative evaluation metrics in protein complexes. Finally, we analyze the formidable challenges faced in current protein complex structure prediction tasks, including the structure prediction of heteromeric complex, disordered regions in complex, antibody-antigen complex, and RNA-related complex, as well as the evaluation metrics for complex assessment. We hope that this work will provide comprehensive knowledge of complex structure predictions to contribute to future advanced predictions.

CombFold: predicting structures of large protein assemblies using a combinatorial assembly algorithm and AlphaFold2

Deep learning models, such as AlphaFold2 and RosettaFold, enable high-accuracy protein structure prediction. However, large protein complexes are still challenging to predict due to their size and the complexity of interactions between multiple subunits. Here we present CombFold, a combinatorial and hierarchical assembly algorithm for predicting structures of large protein complexes utilizing pairwise interactions between subunits predicted by AlphaFold2. CombFold accurately predicted (TM-score >0.7) 72% of the complexes among the top-10 predictions in two datasets of 60 large, asymmetric assemblies. Moreover, the structural coverage of predicted complexes was 20% higher compared to corresponding Protein Data Bank entries. We applied the method on complexes from Complex Portal with known stoichiometry but without known structure and obtained high-confidence predictions. CombFold supports the integration of distance restraints based on crosslinking mass spectrometry and fast enumeration of possible complex stoichiometries. CombFold's high accuracy makes it a promising tool for expanding structural coverage beyond monomeric proteins.

Flexible protein-protein docking with a multitrack iterative transformer

Conventional protein-protein docking algorithms usually rely on heavy candidate sampling and reranking, but these steps are time-consuming and hinder applications that require high-throughput complex structure prediction, for example, structure-based virtual screening. Existing deep learning methods for protein-protein docking, despite being much faster, suffer from low docking success rates. In addition, they simplify the problem to assume no conformational changes within any protein upon binding (rigid docking). This assumption precludes applications when binding-induced conformational changes play a role, such as allosteric inhibition or docking from uncertain unbound model structures. To address these limitations, we present GeoDock, a multitrack iterative transformer network to predict a docked structure from separate docking partners. Unlike deep learning models for protein structure prediction that input multiple sequence alignments, GeoDock inputs just the sequences and structures of the docking partners, which suits the tasks when the individual structures are given. GeoDock is flexible at the protein residue level, allowing the prediction of conformational changes upon binding. On the Database of Interacting Protein Structures (DIPS) test set, GeoDock achieves a 43% top-1 success rate, outperforming all other tested methods. However, in the standard DIPS train/test splits, we discovered contamination of close homologs in the training set. After decontaminating the training set, the success rate is 31%. On the DB5.5 test set and a benchmark dataset of antibody-antigen complexes, GeoDock outperforms the deep learning models trained using the same dataset but falls behind most of the conventional methods and AlphaFold-Multimer. GeoDock attains an average inference speed of under 1 s on a single GPU, enabling its application in large-scale structure screening. Although binding-induced conformational changes are still a challenge owing to limited training and evaluation data, our architecture sets up the foundation to capture this backbone flexibility. Code and a demonstration Jupyter notebook are available at https://github.com/Graylab/GeoDock.

Assessment of Protein-Protein Docking Models Using Deep Learning

Protein-protein interactions are involved in almost all processes in a living cell and determine the biological functions of proteins. To obtain mechanistic understandings of protein-protein interactions, the tertiary structures of protein complexes have been determined by biophysical experimental methods, such as X-ray crystallography and cryogenic electron microscopy. However, as experimental methods are costly in resources, many computational methods have been developed that model protein complex structures. One of the difficulties in computational protein complex modeling (protein docking) is to select the most accurate models among many models that are usually generated by a docking method. This article reviews advances in protein docking model assessment methods, focusing on recent developments that apply deep learning to several network architectures.

Approaches to Backbone Flexibility in Protein-Protein Docking

Proteins are the fundamental organic macromolecules in living systems that play a key role in a variety of biological functions including immunological detection, intracellular trafficking, and signal transduction. The docking of proteins has greatly advanced during recent decades and has become a crucial complement to experimental methods. Protein-protein docking is a helpful method for simulating protein complexes whose structures have not yet been solved experimentally. This chapter focuses on major search tactics along with various docking programs used in protein-protein docking algorithms, which include: direct search, exhaustive global search, local shape feature matching, randomized search, and broad category of post-docking approaches. As backbone flexibility predictions and interactions in high-resolution protein-protein docking remain important issues in the overall optimization context, we have put forward several methods and solutions used to handle backbone flexibility. In addition, various docking methods that are utilized for flexible backbone docking, including ATTRACT, FlexDock, FLIPDock, HADDOCK, RosettaDock, FiberDock, etc., along with their scoring functions, algorithms, advantages, and limitations are discussed. Moreover, what progress in search technology is expected, including not only the creation of new search algorithms but also the enhancement of existing ones, has been debated. As conformational flexibility is one of the most crucial factors affecting docking success, more work should be put into evaluating the conformational flexibility upon binding for a particular case in addition to developing new algorithms to replace the rigid body docking and scoring approach.

From Multiple Protein Docking to Protein-Protein Docking at Interactome Level

Molecular docking is used to anticipate the optimal orientation of a particular molecule to a target to form a stable complex. It makes predictions about the 3D structure of any complex based on the binding characteristics of the ligand and the target receptor usually a protein. It is an exceptionally useful tool, which is used as a model to study how ligands attach to proteins. Docking can also be used for studying the interaction of ligands and proteins to analyze inhibitory efficacy. The ligand may also be a protein, making it possible to study interactions between two different proteins using the numerous docking tools available for basic research on protein interactions. The protein-protein docking is a crucial approach to understanding the protein interactions and predicting the structure of protein complexes that have not yet been experimentally determined. Moreover, the protein-protein interactions can predict the function of target proteins and the drug-like properties of molecules. Therefore, protein docking assists in uncovering insights into protein interactions and also aids in a better understanding of molecular pathways/mechanisms. This chapter comprehends the various tools for protein-protein docking (pairwise and multiple), including their methodologies and analysis of output as results.

Sampling and Scoring in Protein-Protein Docking

Protein-protein docking is considered one of the most important techniques supporting experimental proteomics. Recent developments in the field of computer science helped to improve this computational technique so that it better handles the complexity of protein nature. Sampling algorithms are responsible for the generation of numerous protein-protein ensembles. Unfortunately, a primary docking output comprises a set of both near-native poses and decoys. Application of the efficient scoring function helps to differentiate poses with the most favorable properties from those that are very unlikely to represent a natural state of the complex. This chapter explains the importance of sampling and scoring in the process of protein-protein docking. Moreover, it summarizes advances in the field.

Discriminating physiological from non-physiological interfaces in structures of protein complexes: A community-wide study

Reliably scoring and ranking candidate models of protein complexes and assigning their oligomeric state from the structure of the crystal lattice represent outstanding challenges. A community-wide effort was launched to tackle these challenges. The latest resources on protein complexes and interfaces were exploited to derive a benchmark dataset consisting of 1677 homodimer protein crystal structures, including a balanced mix of physiological and non-physiological complexes. The non-physiological complexes in the benchmark were selected to bury a similar or larger interface area than their physiological counterparts, making it more difficult for scoring functions to differentiate between them. Next, 252 functions for scoring protein-protein interfaces previously developed by 13 groups were collected and evaluated for their ability to discriminate between physiological and non-physiological complexes. A simple consensus score generated using the best performing score of each of the 13 groups, and a cross-validated Random Forest (RF) classifier were created. Both approaches showed excellent performance, with an area under the Receiver Operating Characteristic (ROC) curve of 0.93 and 0.94, respectively, outperforming individual scores developed by different groups. Additionally, AlphaFold2 engines recalled the physiological dimers with significantly higher accuracy than the non-physiological set, lending support to the reliability of our benchmark dataset annotations. Optimizing the combined power of interface scoring functions and evaluating it on challenging benchmark datasets appears to be a promising strategy.

Flexible Protein-Protein Docking with a Multi-Track Iterative Transformer

Conventional protein-protein docking algorithms usually rely on heavy candidate sampling and re-ranking, but these steps are time-consuming and hinder applications that require high-throughput complex structure prediction, e.g., structure-based virtual screening. Existing deep learning methods for protein-protein docking, despite being much faster, suffer from low docking success rates. In addition, they simplify the problem to assume no conformational changes within any protein upon binding (rigid docking). This assumption precludes applications when binding-induced conformational changes play a role, such as allosteric inhibition or docking from uncertain unbound model structures. To address these limitations, we present GeoDock, a multi-track iterative transformer network to predict a docked structure from separate docking partners. Unlike deep learning models for protein structure prediction that input multiple sequence alignments (MSAs), GeoDock inputs just the sequences and structures of the docking partners, which suits the tasks when the individual structures are given. GeoDock is flexible at the protein residue level, allowing the prediction of conformational changes upon binding. For a benchmark set of rigid targets, GeoDock obtains a 41% success rate, outperforming all the other tested methods. For a more challenging benchmark set of flexible targets, GeoDock achieves a similar number of top-model successes as the traditional method ClusPro [1], but fewer than ReplicaDock2 [2]. GeoDock attains an average inference speed of under one second on a single GPU, enabling its application in large-scale structure screening. Although binding-induced conformational changes are still a challenge owing to limited training and evaluation data, our architecture sets up the foundation to capture this backbone flexibility. Code and a demonstration Jupyter notebook are available at https://github.com/Graylab/GeoDock.

Evaluation of AlphaFold-Multimer prediction on multi-chain protein complexes

MOTIVATION

Despite near-experimental accuracy on single-chain predictions, there is still scope for improvement among multimeric predictions. Methods like AlphaFold-Multimer and FoldDock can accurately model dimers. However, how well these methods fare on larger complexes is still unclear. Further, evaluation methods of the quality of multimeric complexes are not well established.

RESULTS

We analysed the performance of AlphaFold-Multimer on a homology-reduced dataset of homo- and heteromeric protein complexes. We highlight the differences between the pairwise and multi-interface evaluation of chains within a multimer. We describe why certain complexes perform well on one metric (e.g. TM-score) but poorly on another (e.g. DockQ). We propose a new score, Predicted DockQ version 2 (pDockQ2), to estimate the quality of each interface in a multimer. Finally, we modelled protein complexes (from CORUM) and identified two highly confident structures that do not have sequence homology to any existing structures.

AVAILABILITY AND IMPLEMENTATION

All scripts, models, and data used to perform the analysis in this study are freely available at https://gitlab.com/ElofssonLab/afm-benchmark.

Predicting structures of large protein assemblies using combinatorial assembly algorithm and AlphaFold2

Deep learning models, such as AlphaFold2 and RosettaFold, enable high-accuracy protein structure prediction. However, large protein complexes are still challenging to predict due to their size and the complexity of interactions between multiple subunits. Here we present CombFold, a combinatorial and hierarchical assembly algorithm for predicting structures of large protein complexes utilizing pairwise interactions between subunits predicted by AlphaFold2. CombFold accurately predicted (TM-score > 0.7) 72% of the complexes among the Top-10 predictions in two datasets of 60 large, asymmetric assemblies. Moreover, the structural coverage of predicted complexes was 20% higher compared to corresponding PDB entries. We applied the method on complexes from Complex Portal with known stoichiometry but without known structure and obtained high-confidence predictions. CombFold supports the integration of distance restraints based on crosslinking mass spectrometry and fast enumeration of possible complex stoichiometries. CombFold's high accuracy makes it a promising tool for expanding structural coverage beyond monomeric proteins.

Computational Methods and Deep Learning for Elucidating Protein Interaction Networks

Protein interactions play a critical role in all biological processes, but experimental identification of protein interactions is a time- and resource-intensive process. The advances in next-generation sequencing and multi-omics technologies have greatly benefited large-scale predictions of protein interactions using machine learning methods. A wide range of tools have been developed to predict protein-protein, protein-nucleic acid, and protein-drug interactions. Here, we discuss the applications, methods, and challenges faced when employing the various prediction methods. We also briefly describe ways to overcome the challenges and prospective future developments in the field of protein interaction biology.

Pairwise and Multi-chain Protein Docking Enhanced Using LZerD Web Server

Interactions of proteins with other macromolecules have important structural and functional roles in the basic processes of living cells. To understand and elucidate the mechanisms of interactions, it is important to know the 3D structures of the complexes. Proteomes contain numerous protein-protein complexes, for which experimentally determined structures often do not exist. Computational techniques can be a practical alternative to obtain useful complex structure models. Here, we present a web server that provides access to the LZerD and Multi-LZerD protein docking tools, which can perform both pairwise and multi-chain docking. The web server is user-friendly, with options to visualize the distribution and structures of binding poses of top-scoring models. The LZerD web server is available at https://lzerd.kiharalab.org . This chapter dictates the algorithm and step-by-step procedure to model the monomeric structures with AttentiveDist, and also provides the detail of pairwise LZerD docking, and multi-LZerD. This also provided case studies for each of the three modules.

Domain-Based Protein Docking with Extremely Large Conformational Changes

Proteins are key components in many processes in living cells, and physical interactions with other proteins and nucleic acids often form key parts of their functions. In many cases, large flexibility of proteins as they interact is key to their function. To understand the mechanisms of these processes, it is necessary to consider the 3D structures of such protein complexes. When such structures are not yet experimentally determined, protein docking has long been present to computationally generate useful structure models. However, protein docking has long had the limitation that the consideration of flexibility is usually limited to very small movements or very small structures. Methods have been developed which handle minor flexibility via normal mode or other structure sampling, but new methods are required to model ordered proteins which undergo large-scale conformational changes to elucidate their function at the molecular level. Here, we present Flex-LZerD, a framework for docking such complexes. Via partial assembly multidomain docking and an iterative normal mode analysis admitting curvilinear motions, we demonstrate the ability to model the assembly of a variety of protein-protein and protein-nucleic acid complexes.

Predicting the structure of large protein complexes using AlphaFold and Monte Carlo tree search

AlphaFold can predict the structure of single- and multiple-chain proteins with very high accuracy. However, the accuracy decreases with the number of chains, and the available GPU memory limits the size of protein complexes which can be predicted. Here we show that one can predict the structure of large complexes starting from predictions of subcomponents. We assemble 91 out of 175 complexes with 10-30 chains from predicted subcomponents using Monte Carlo tree search, with a median TM-score of 0.51. There are 30 highly accurate complexes (TM-score ≥0.8, 33% of complete assemblies). We create a scoring function, mpDockQ, that can distinguish if assemblies are complete and predict their accuracy. We find that complexes containing symmetry are accurately assembled, while asymmetrical complexes remain challenging. The method is freely available and accesible as a Colab notebook https://colab.research.google.com/github/patrickbryant1/MoLPC/blob/master/MoLPC.ipynb .

RL-MLZerD: Multimeric protein docking using reinforcement learning

Numerous biological processes in a cell are carried out by protein complexes. To understand the molecular mechanisms of such processes, it is crucial to know the quaternary structures of the complexes. Although the structures of protein complexes have been determined by biophysical experiments at a rapid pace, there are still many important complex structures that are yet to be determined. To supplement experimental structure determination of complexes, many computational protein docking methods have been developed; however, most of these docking methods are designed only for docking with two chains. Here, we introduce a novel method, RL-MLZerD, which builds multiple protein complexes using reinforcement learning (RL). In RL-MLZerD a multi-chain assembly process is considered as a series of episodes of selecting and integrating pre-computed pairwise docking models in a RL framework. RL is effective in correctly selecting plausible pairwise models that fit well with other subunits in a complex. When tested on a benchmark dataset of protein complexes with three to five chains, RL-MLZerD showed better modeling performance than other existing multiple docking methods under different evaluation criteria, except against AlphaFold-Multimer in unbound docking. Also, it emerged that the docking order of multi-chain complexes can be naturally predicted by examining preferred paths of episodes in the RL computation.

Efficient DNA-ligand interaction framework using fuzzy C-means clustering based glowworm swarm optimization (FCMGSO) method

Assessment of DNA and ligand interaction is a great challenge to the medical researchers and drug industries since the accurate mapping of DNA and ligand plays an important role in associating drugs for suitable diseases. The primary objective of this research work is to develop an efficient model for predicting the best DNA and Ligand mapping. In this research work, 500 instances of DNA and drugs used for cancer and non-cancer diseases from the National Centre for Biotechnology Information (NCBI) were considered for analysis. Binding energy is one of the important measures to predict and finalize the best DNA and ligand interaction. Existing methods used for the docking process such as Simulated Annealing (SA), Lamarckian Genetic Algorithm (LGA), Genetic Clustering (GC), Fuzzy C-means clustering (FCM), and Genetic Clustering with Multi swarm Optimization (GCMSO) were applied for all 500 instances. These algorithms failed to produce better binding energy due to a lack of optimization in the existing approaches. Optimization methods play a major role in predicting accurate DNA ligand docking. Hence, this research proposes an efficient architecture using Fuzzy C-Means Clustering with Glowworm Swarm (FCMGSO) optimization method for accurate analysis of the DNA-ligand docking process. Results are proving that the proposed FCMGSO algorithm shows less binding energy than other existing methods in all instances of samples considered from the NCBI dataset.Communicated by Ramaswamy H. Sarma.

Multi-level analysis of intrinsically disordered protein docking methods

Intrinsically Disordered Proteins (IDPs) are a class of proteins in which at least some region of the protein does not possess any stable structure in solution in the physiological condition but may adopt an ordered structure upon binding to a globular receptor. These IDP-receptor complexes are thus subject to protein complex modeling in which computational techniques are applied to accurately reproduce the IDP ligand-receptor interactions. This often exists in the form of protein docking, in which the 3D structures of both the subunits are known, but the position of the ligand relative to the receptor is not. Here, we evaluate the performance of three IDP-receptor modeling tools with metrics that characterize the IDP-receptor interface at various resolutions. We show that all three methods are able to properly identify the general binding site, as identified by lower resolution metrics, but begin to struggle with higher resolution metrics that capture biophysical interactions.

MarkovFit: Structure Fitting for Protein Complexes in Electron Microscopy Maps Using Markov Random Field

An increasing number of protein complex structures are determined by cryo-electron microscopy (cryo-EM). When individual protein structures have been determined and are available, an important task in structure modeling is to fit the individual structures into the density map. Here, we designed a method that fits the atomic structures of proteins in cryo-EM maps of medium to low resolutions using Markov random fields, which allows probabilistic evaluation of fitted models. The accuracy of our method, MarkovFit, performed better than existing methods on datasets of 31 simulated cryo-EM maps of resolution 10 Å , nine experimentally determined cryo-EM maps of resolution less than 4 Å , and 28 experimentally determined cryo-EM maps of resolution 6 to 20 Å .

Prediction of protein assemblies, the next frontier: The CASP14-CAPRI experiment

We present the results for CAPRI Round 50, the fourth joint CASP-CAPRI protein assembly prediction challenge. The Round comprised a total of twelve targets, including six dimers, three trimers, and three higher-order oligomers. Four of these were easy targets, for which good structural templates were available either for the full assembly, or for the main interfaces (of the higher-order oligomers). Eight were difficult targets for which only distantly related templates were found for the individual subunits. Twenty-five CAPRI groups including eight automatic servers submitted ~1250 models per target. Twenty groups including six servers participated in the CAPRI scoring challenge submitted ~190 models per target. The accuracy of the predicted models was evaluated using the classical CAPRI criteria. The prediction performance was measured by a weighted scoring scheme that takes into account the number of models of acceptable quality or higher submitted by each group as part of their five top-ranking models. Compared to the previous CASP-CAPRI challenge, top performing groups submitted such models for a larger fraction (70-75%) of the targets in this Round, but fewer of these models were of high accuracy. Scorer groups achieved stronger performance with more groups submitting correct models for 70-80% of the targets or achieving high accuracy predictions. Servers performed less well in general, except for the MDOCKPP and LZERD servers, who performed on par with human groups. In addition to these results, major advances in methodology are discussed, providing an informative overview of where the prediction of protein assemblies currently stands.

LZerD Protein-Protein Docking Webserver Enhanced With de novo Structure Prediction

Protein-protein docking is a useful tool for modeling the structures of protein complexes that have yet to be experimentally determined. Understanding the structures of protein complexes is a key component for formulating hypotheses in biophysics regarding the functional mechanisms of complexes. Protein-protein docking is an established technique for cases where the structures of the subunits have been determined. While the number of known structures deposited in the Protein Data Bank is increasing, there are still many cases where the structures of individual proteins that users want to dock are not determined yet. Here, we have integrated the AttentiveDist method for protein structure prediction into our LZerD webserver for protein-protein docking, which enables users to simply submit protein sequences and obtain full-complex atomic models, without having to supply any structure themselves. We have further extended the LZerD docking interface with a symmetrical homodimer mode. The LZerD server is available at https://lzerd.kiharalab.org/.

LZerD webserver for pairwise and multiple protein-protein docking

Protein complexes are involved in many important processes in living cells. To understand the mechanisms of these processes, it is necessary to solve the 3D structures of the protein complexes. When protein complex structures have not yet been determined by experiment, protein-protein docking tools can be used to computationally model the structures of these complexes. Here, we present a webserver which provides access to LZerD and Multi-LZerD protein docking tools. The protocol provided by the server have performed consistently among the top in the CAPRI blind evaluation. LZerD docks pairs of structures, while Multi-LZerD can dock three or more structures simultaneously. LZerD uses a soft protein surface representation with 3D Zernike descriptors and explores the binding pose space using geometric hashing. Multi-LZerD performs multi-chain docking by combining pairwise solutions by LZerD. Both methods output full-atom docked models of the input proteins. Users can also input distance constraints between interacting or non-interacting residues as well as residues that locate at the interface or far from the interface. The webserver is equipped with a user-friendly panel that visualizes the distribution and structures of binding poses of top scoring models. The LZerD webserver is available at https://lzerd.kiharalab.org.

Modeling Perturbations in Protein Filaments at the Micro and Meso Scale Using NAMD and PTools/Heligeom

Protein filaments are dynamic entities that respond to external stimuli by slightly or substantially modifying the internal binding geometries between successive protomers. This results in overall changes in the filament architecture, which are difficult to model due to the helical character of the system. Here, we describe how distortions in RecA nucleofilaments and their consequences on the filament-DNA and bound DNA-DNA interactions at different stages of the homologous recombination process can be modeled using the PTools/Heligeom software and subsequent molecular dynamics simulation with NAMD. Modeling methods dealing with helical macromolecular objects typically rely on symmetric assemblies and take advantage of known symmetry descriptors. Other methods dealing with single objects, such as MMTK or VMD, do not integrate the specificities of regular assemblies. By basing the model building on binding geometries at the protomer-protomer level, PTools/Heligeom frees the building process from a priori knowledge of the system topology and enables irregular architectures and symmetry disruption to be accounted for. Graphical abstract: Model of ATP hydrolysis-induced distortions in the recombinant nucleoprotein, obtained by combining RecA-DNA and two RecA-RecA binding geometries.

EvoSeg: Automated Electron Microscopy Segmentation through Random Forests and Evolutionary Optimization

Electron Microscopy Maps are key in the study of bio-molecular structures, ranging from borderline atomic level to the sub-cellular range. These maps describe the envelopes that cover possibly a very large number of proteins that form molecular machines within the cell. Within those envelopes, we are interested to find what regions correspond to specific proteins so that we can understand how they function, and design drugs that can enhance or suppress a process that they are involved in, along with other experimental purposes. A classic approach by which we can begin the exploration of map regions is to apply a segmentation algorithm. This yields a mask where each voxel in 3D space is assigned an identifier that maps it to a segment; an ideal segmentation would map each segment to one protein unit, which is rarely the case. In this work, we present a method that uses bio-inspired optimization, through an Evolutionary-Optimized Segmentation algorithm, to iteratively improve upon baseline segments obtained from a classical approach, called watershed segmentation. The cost function used by the evolutionary optimization is based on an ideal segmentation classifier trained as part of this development, which uses basic structural information available to scientists, such as the number of expected units, volume and topology. We show that a basic initial segmentation with the additional information allows our evolutionary method to find better segmentation results, compared to the baseline generated by the watershed.

Protein Docking Model Evaluation by Graph Neural Networks

Physical interactions of proteins play key functional roles in many important cellular processes. To understand molecular mechanisms of such functions, it is crucial to determine the structure of protein complexes. To complement experimental approaches, which usually take a considerable amount of time and resources, various computational methods have been developed for predicting the structures of protein complexes. In computational modeling, one of the challenges is to identify near-native structures from a large pool of generated models. Here, we developed a deep learning-based approach named Graph Neural Network-based DOcking decoy eValuation scorE (GNN-DOVE). To evaluate a protein docking model, GNN-DOVE extracts the interface area and represents it as a graph. The chemical properties of atoms and the inter-atom distances are used as features of nodes and edges in the graph, respectively. GNN-DOVE was trained, validated, and tested on docking models in the Dockground database and further tested on a combined dataset of Dockground and ZDOCK benchmark as well as a CAPRI scoring dataset. GNN-DOVE performed better than existing methods, including DOVE, which is our previous development that uses a convolutional neural network on voxelized structure models.

Computational structure modeling for diverse categories of macromolecular interactions

Computational protein-protein docking is one of the most intensively studied topics in structural bioinformatics. The field has made substantial progress through over three decades of development. The development began with methods for rigid-body docking of two proteins, which have now been extended in different directions to cover the various macromolecular interactions observed in a cell. Here, we overview the recent developments of the variations of docking methods, including multiple protein docking, peptide-protein docking, and disordered protein docking methods.

Protein docking model evaluation by 3D deep convolutional neural networks

MOTIVATION

Many important cellular processes involve physical interactions of proteins. Therefore, determining protein quaternary structures provide critical insights for understanding molecular mechanisms of functions of the complexes. To complement experimental methods, many computational methods have been developed to predict structures of protein complexes. One of the challenges in computational protein complex structure prediction is to identify near-native models from a large pool of generated models.

RESULTS

We developed a convolutional deep neural network-based approach named DOcking decoy selection with Voxel-based deep neural nEtwork (DOVE) for evaluating protein docking models. To evaluate a protein docking model, DOVE scans the protein-protein interface of the model with a 3D voxel and considers atomic interaction types and their energetic contributions as input features applied to the neural network. The deep learning models were trained and validated on docking models available in the ZDock and DockGround databases. Among the different combinations of features tested, almost all outperformed existing scoring functions.

AVAILABILITY AND IMPLEMENTATION

Codes available at http://github.com/kiharalab/DOVE, http://kiharalab.org/dove/.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Quantitative Characterization of Binding Pockets and Binding Complementarity by Means of Zernike Descriptors

In this work, we describe the application of the Zernike formalism to quantitatively characterize the binding pockets of two sets of biologically relevant systems. Such an approach, when applied to molecular dynamics trajectories, is able to pinpoint the subtle differences between very similar molecular regions and their impact on the local propensity to ligand binding, allowing us to quantify such differences. The statistical robustness of our procedure suggests that it is very suitable to describe protein binding sites and protein-ligand interactions within a rigorous and well-defined framework.

Modeling of Multimolecular Complexes

Macromolecular complexes play a key role in cellular function. Predicting the structure and dynamics of these complexes is one of the key challenges in structural biology. Docking applications have traditionally been used to predict pairwise interactions between proteins. However, few methods exist for modeling multi-protein assemblies. Here we present two methods, CombDock and DockStar, that can predict multi-protein assemblies starting from subunit structural models. CombDock can assemble subunits without any assumptions about the pairwise interactions between subunits, while DockStar relies on the interaction graph or, alternatively, a homology model or a cryo-electron microscopy (EM) density map of the entire complex. We demonstrate the two methods using RNA polymerase II with 12 subunits and TRiC/CCT chaperonin with 16 subunits.

Multiscale Visual Drilldown for the Analysis of Large Ensembles of Multi-Body Protein Complexes

When studying multi-body protein complexes, biochemists use computational tools that can suggest hundreds or thousands of their possible spatial configurations. However, it is not feasible to experimentally verify more than only a very small subset of them. In this paper, we propose a novel multiscale visual drilldown approach that was designed in tight collaboration with proteomic experts, enabling a systematic exploration of the configuration space. Our approach takes advantage of the hierarchical structure of the data - from the whole ensemble of protein complex configurations to the individual configurations, their contact interfaces, and the interacting amino acids. Our new solution is based on interactively linked 2D and 3D views for individual hierarchy levels. At each level, we offer a set of selection and filtering operations that enable the user to narrow down the number of configurations that need to be manually scrutinized. Furthermore, we offer a dedicated filter interface, which provides the users with an overview of the applied filtering operations and enables them to examine their impact on the explored ensemble. This way, we maintain the history of the exploration process and thus enable the user to return to an earlier point of the exploration. We demonstrate the effectiveness of our approach on two case studies conducted by collaborating proteomic experts.

Blind prediction of homo- and hetero-protein complexes: The CASP13-CAPRI experiment

We present the results for CAPRI Round 46, the third joint CASP-CAPRI protein assembly prediction challenge. The Round comprised a total of 20 targets including 14 homo-oligomers and 6 heterocomplexes. Eight of the homo-oligomer targets and one heterodimer comprised proteins that could be readily modeled using templates from the Protein Data Bank, often available for the full assembly. The remaining 11 targets comprised 5 homodimers, 3 heterodimers, and two higher-order assemblies. These were more difficult to model, as their prediction mainly involved "ab-initio" docking of subunit models derived from distantly related templates. A total of ~30 CAPRI groups, including 9 automatic servers, submitted on average ~2000 models per target. About 17 groups participated in the CAPRI scoring rounds, offered for most targets, submitting ~170 models per target. The prediction performance, measured by the fraction of models of acceptable quality or higher submitted across all predictors groups, was very good to excellent for the nine easy targets. Poorer performance was achieved by predictors for the 11 difficult targets, with medium and high quality models submitted for only 3 of these targets. A similar performance "gap" was displayed by scorer groups, highlighting yet again the unmet challenge of modeling the conformational changes of the protein components that occur upon binding or that must be accounted for in template-based modeling. Our analysis also indicates that residues in binding interfaces were less well predicted in this set of targets than in previous Rounds, providing useful insights for directions of future improvements.

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.

Predicting binding poses and affinity ranking in D3R Grand Challenge using PL-PatchSurfer2.0

Computational prediction of protein-ligand interactions is a useful approach that aids the drug discovery process. Two major tasks of computational approaches are to predict the docking pose of a compound in a known binding pocket and to rank compounds in a library according to their predicted binding affinities. There are many computational tools developed in the past decades both in academia and industry. To objectively assess the performance of existing tools, the community has held a blind assessment of computational predictions, the Drug Design Data Resource Grand Challenge. This round, Grand Challenge 4 (GC4), focused on two targets, protein beta-secretase 1 (BACE-1) and cathepsin S (CatS). We participated in GC4 in both BACE-1 and CatS challenges using our molecular surface-based virtual screening method, PL-PatchSurfer2.0. A unique feature of PL-PatchSurfer2.0 is that it uses the three-dimensional Zernike descriptor, a mathematical moment-based shape descriptor, to quantify local shape complementarity between a ligand and a receptor, which properly incorporates molecular flexibility and provides stable affinity assessment for a bound ligand-receptor complex. Since PL-PatchSurfer2.0 does not explicitly build a bound pose of a ligand, we used an external docking program, such as AutoDock Vina, to provide an ensemble of poses, which were then evaluated by PL-PatchSurfer2.0. Here, we provide an overview of our method and report the performance in GC4.

Analysis of Protein Complexes in the Unicellular Cyanobacterium Cyanothece ATCC 51142

The unicellular cyanobacterium Cyanothece ATCC 51142 is capable of oxygenic photosynthesis and biological N₂ fixation (BNF), a process highly sensitive to oxygen. Previous work has focused on determining protein expression levels under different growth conditions. A major gap of our knowledge is an understanding on how these expressed proteins are assembled into complexes and organized into metabolic pathways, an area that has not been thoroughly investigated. Here, we combined size-exclusion chromatography (SEC) with label-free quantitative mass spectrometry (MS) and bioinformatics to characterize many protein complexes from Cyanothece 51142 cells grown under a 12 h light-dark cycle. We identified 1386 proteins in duplicate biological replicates, and 64% of those proteins were identified as putative complexes. Pairwise computational prediction of protein-protein interaction (PPI) identified 74 822 putative interactions, of which 2337 interactions were highly correlated with published protein coexpressions. Many sequential glycolytic and TCA cycle enzymes were identified as putative complexes. We also identified many membrane complexes that contain cytoplasmic domains. Subunits of NDH-1 complex eluted in a fraction with an approximate mass of ∼669 kDa, and subunits composition revealed coexistence of distinct forms of NDH-1 complex subunits responsible for respiration, electron flow, and CO₂ uptake. The complex form of the phycocyanin beta subunit was nonphosphorylated, and the monomer form was phosphorylated at Ser20, suggesting phosphorylation-dependent deoligomerization of the phycocyanin beta subunit. This study provides an analytical platform for future studies to reveal how these complexes assemble and disassemble as a function of diurnal and circadian rhythms.

Computational Methods for Predicting Protein-Protein Interactions Using Various Protein Features

Understanding protein-protein interactions (PPIs) in a cell is essential for learning protein functions, pathways, and mechanism of diseases. PPIs are also important targets for developing drugs. Experimental methods, both small-scale and large-scale, have identified PPIs in several model organisms. However, results cover only a part of PPIs of organisms; moreover, there are many organisms whose PPIs have not yet been investigated. To complement experimental methods, many computational methods have been developed that predict PPIs from various characteristics of proteins. Here we provide an overview of literature reports to classify computational PPI prediction methods that consider different features of proteins, including protein sequence, genomes, protein structure, function, PPI network topology, and those which integrate multiple methods. © 2018 by John Wiley & Sons, Inc.

From Optimization to Mapping: An Evolutionary Algorithm for Protein Energy Landscapes

Stochastic search is often the only viable option to address complex optimization problems. Recently, evolutionary algorithms have been shown to handle challenging continuous optimization problems related to protein structure modeling. Building on recent work in our laboratories, we propose an evolutionary algorithm for efficiently mapping the multi-basin energy landscapes of dynamic proteins that switch between thermodynamically stable or semi-stable structural states to regulate their biological activity in the cell. The proposed algorithm balances computational resources between exploration and exploitation of the nonlinear, multimodal landscapes that characterize multi-state proteins via a novel combination of global and local search to generate a dynamically-updated, information-rich map of a protein's energy landscape. This new mapping-oriented EA is applied to several dynamic proteins and their disease-implicated variants to illustrate its ability to map complex energy landscapes in a computationally feasible manner. We further show that, given the availability of such maps, comparison between the maps of wildtype and variants of a protein allows for the formulation of a structural and thermodynamic basis for the impact of sequence mutations on dysfunction that may prove useful in guiding further wet-laboratory investigations of dysfunction and molecular interventions.

Improved performance in CAPRI round 37 using LZerD docking and template-based modeling with combined scoring functions

We report our group's performance for protein-protein complex structure prediction and scoring in Round 37 of the Critical Assessment of PRediction of Interactions (CAPRI), an objective assessment of protein-protein complex modeling. We demonstrated noticeable improvement in both prediction and scoring compared to previous rounds of CAPRI, with our human predictor group near the top of the rankings and our server scorer group at the top. This is the first time in CAPRI that a server has been the top scorer group. To predict protein-protein complex structures, we used both multi-chain template-based modeling (TBM) and our protein-protein docking program, LZerD. LZerD represents protein surfaces using 3D Zernike descriptors (3DZD), which are based on a mathematical series expansion of a 3D function. Because 3DZD are a soft representation of the protein surface, LZerD is tolerant to small conformational changes, making it well suited to docking unbound and TBM structures. The key to our improved performance in CAPRI Round 37 was to combine multi-chain TBM and docking. As opposed to our previous strategy of performing docking for all target complexes, we used TBM when multi-chain templates were available and docking otherwise. We also describe the combination of multiple scoring functions used by our server scorer group, which achieved the top rank for the scorer phase.

Modeling the assembly order of multimeric heteroprotein complexes

Protein-protein interactions are the cornerstone of numerous biological processes. Although an increasing number of protein complex structures have been determined using experimental methods, relatively fewer studies have been performed to determine the assembly order of complexes. In addition to the insights into the molecular mechanisms of biological function provided by the structure of a complex, knowing the assembly order is important for understanding the process of complex formation. Assembly order is also practically useful for constructing subcomplexes as a step toward solving the entire complex experimentally, designing artificial protein complexes, and developing drugs that interrupt a critical step in the complex assembly. There are several experimental methods for determining the assembly order of complexes; however, these techniques are resource-intensive. Here, we present a computational method that predicts the assembly order of protein complexes by building the complex structure. The method, named Path-LzerD, uses a multimeric protein docking algorithm that assembles a protein complex structure from individual subunit structures and predicts assembly order by observing the simulated assembly process of the complex. Benchmarked on a dataset of complexes with experimental evidence of assembly order, Path-LZerD was successful in predicting the assembly pathway for the majority of the cases. Moreover, when compared with a simple approach that infers the assembly path from the buried surface area of subunits in the native complex, Path-LZerD has the strong advantage that it can be used for cases where the complex structure is not known. The path prediction accuracy decreased when starting from unbound monomers, particularly for larger complexes of five or more subunits, for which only a part of the assembly path was correctly identified. As the first method of its kind, Path-LZerD opens a new area of computational protein structure modeling and will be an indispensable approach for studying protein complexes.

Protein 3D Structure and Electron Microscopy Map Retrieval Using 3D-SURFER2.0 and EM-SURFER

With the rapid growth in the number of solved protein structures stored in the Protein Data Bank (PDB) and the Electron Microscopy Data Bank (EMDB), it is essential to develop tools to perform real-time structure similarity searches against the entire structure database. Since conventional structure alignment methods need to sample different orientations of proteins in the three-dimensional space, they are time consuming and unsuitable for rapid, real-time database searches. To this end, we have developed 3D-SURFER and EM-SURFER, which utilize 3D Zernike descriptors (3DZD) to conduct high-throughput protein structure comparison, visualization, and analysis. Taking an atomic structure or an electron microscopy map of a protein or a protein complex as input, the 3DZD of a query protein is computed and compared with the 3DZD of all other proteins in PDB or EMDB. In addition, local geometrical characteristics of a query protein can be analyzed using VisGrid and LIGSITE^CSC in 3D-SURFER. This article describes how to use 3D-SURFER and EM-SURFER to carry out protein surface shape similarity searches, local geometric feature analysis, and interpretation of the search results. © 2017 by John Wiley & Sons, Inc.

Predictive and Experimental Approaches for Elucidating Protein-Protein Interactions and Quaternary Structures

The elucidation of protein-protein interactions is vital for determining the function and action of quaternary protein structures. Here, we discuss the difficulty and importance of establishing protein quaternary structure and review in vitro and in silico methods for doing so. Determining the interacting partner proteins of predicted protein structures is very time-consuming when using in vitro methods, this can be somewhat alleviated by use of predictive methods. However, developing reliably accurate predictive tools has proved to be difficult. We review the current state of the art in predictive protein interaction software and discuss the problem of scoring and therefore ranking predictions. Current community-based predictive exercises are discussed in relation to the growth of protein interaction prediction as an area within these exercises. We suggest a fusion of experimental and predictive methods that make use of sparse experimental data to determine higher resolution predicted protein interactions as being necessary to drive forward development.

In silico structure-based approaches to discover protein-protein interaction-targeting drugs

A core concept behind modern drug discovery is finding a small molecule that modulates a function of a target protein. This concept has been successfully applied since the mid-1970s. However, the efficiency of drug discovery is decreasing because the druggable target space in the human proteome is limited. Recently, protein-protein interaction (PPI) has been identified asan emerging target space for drug discovery. PPI plays a pivotal role in biological pathways including diseases. Current human interactome research suggests that the number of PPIs is between 130,000 and 650,000, and only a small number of them have been targeted as drug targets. For traditional drug targets, in silico structure-based methods have been successful in many cases. However, their performance suffers on PPI interfaces because PPI interfaces are different in five major aspects: From a geometric standpoint, they have relatively large interface regions, flat geometry, and the interface surface shape tends to fluctuate upon binding. Also, their interactions are dominated by hydrophobic atoms, which is different from traditional binding-pocket-targeted drugs. Finally, PPI targets usually lack natural molecules that bind to the target PPI interface. Here, we first summarize characteristics of PPI interfaces and their known binders. Then, we will review existing in silico structure-based approaches for discovering small molecules that bind to PPI interfaces.

Modeling protein quaternary structure of homo- and hetero-oligomers beyond binary interactions by homology

Cellular processes often depend on interactions between proteins and the formation of macromolecular complexes. The impairment of such interactions can lead to deregulation of pathways resulting in disease states, and it is hence crucial to gain insights into the nature of macromolecular assemblies. Detailed structural knowledge about complexes and protein-protein interactions is growing, but experimentally determined three-dimensional multimeric assemblies are outnumbered by complexes supported by non-structural experimental evidence. Here, we aim to fill this gap by modeling multimeric structures by homology, only using amino acid sequences to infer the stoichiometry and the overall structure of the assembly. We ask which properties of proteins within a family can assist in the prediction of correct quaternary structure. Specifically, we introduce a description of protein-protein interface conservation as a function of evolutionary distance to reduce the noise in deep multiple sequence alignments. We also define a distance measure to structurally compare homologous multimeric protein complexes. This allows us to hierarchically cluster protein structures and quantify the diversity of alternative biological assemblies known today. We find that a combination of conservation scores, structural clustering, and classical interface descriptors, can improve the selection of homologous protein templates leading to reliable models of protein complexes.

Modeling disordered protein interactions from biophysical principles

Disordered protein-protein interactions (PPIs), those involving a folded protein and an intrinsically disordered protein (IDP), are prevalent in the cell, including important signaling and regulatory pathways. IDPs do not adopt a single dominant structure in isolation but often become ordered upon binding. To aid understanding of the molecular mechanisms of disordered PPIs, it is crucial to obtain the tertiary structure of the PPIs. However, experimental methods have difficulty in solving disordered PPIs and existing protein-protein and protein-peptide docking methods are not able to model them. Here we present a novel computational method, IDP-LZerD, which models the conformation of a disordered PPI by considering the biophysical binding mechanism of an IDP to a structured protein, whereby a local segment of the IDP initiates the interaction and subsequently the remaining IDP regions explore and coalesce around the initial binding site. On a dataset of 22 disordered PPIs with IDPs up to 69 amino acids, successful predictions were made for 21 bound and 18 unbound receptors. The successful modeling provides additional support for biophysical principles. Moreover, the new technique significantly expands the capability of protein structure modeling and provides crucial insights into the molecular mechanisms of disordered PPIs.

Human and server docking prediction for CAPRI round 30-35 using LZerD with combined scoring functions

We report the performance of protein-protein docking predictions by our group for recent rounds of the Critical Assessment of Prediction of Interactions (CAPRI), a community-wide assessment of state-of-the-art docking methods. Our prediction procedure uses a protein-protein docking program named LZerD developed in our group. LZerD represents a protein surface with 3D Zernike descriptors (3DZD), which are based on a mathematical series expansion of a 3D function. The appropriate soft representation of protein surface with 3DZD makes the method more tolerant to conformational change of proteins upon docking, which adds an advantage for unbound docking. Docking was guided by interface residue prediction performed with BindML and cons-PPISP as well as literature information when available. The generated docking models were ranked by a combination of scoring functions, including PRESCO, which evaluates the native-likeness of residues' spatial environments in structure models. First, we discuss the overall performance of our group in the CAPRI prediction rounds and investigate the reasons for unsuccessful cases. Then, we examine the performance of several knowledge-based scoring functions and their combinations for ranking docking models. It was found that the quality of a pool of docking models generated by LZerD, that is whether or not the pool includes near-native models, can be predicted by the correlation of multiple scores. Although the current analysis used docking models generated by LZerD, findings on scoring functions are expected to be universally applicable to other docking methods. Proteins 2017; 85:513-527. © 2016 Wiley Periodicals, Inc.

Ranking protein-protein docking results using steered molecular dynamics and potential of mean force calculations

Crystallization of protein-protein complexes can often be problematic and therefore computational structural models are often relied on. Such models are often generated using protein-protein docking algorithms, where one of the main challenges is selecting which of several thousand potential predictions represents the most near-native complex. We have developed a novel technique that involves the use of steered molecular dynamics (sMD) and umbrella sampling to identify near-native complexes among protein-protein docking predictions. Using this technique, we have found a strong correlation between our predictions and the interface RMSD (iRMSD) in ten diverse test systems. On two of the systems, we investigated if the prediction results could be further improved using potential of mean force calculations. We demonstrated that a near-native (<2.0 Å iRMSD) structure could be identified in the top-1 ranked position for both systems. © 2016 Wiley Periodicals, Inc.