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

Ligand Gaussian accelerated Molecular Dynamics 3 (LiGaMD3): Improved Calculations of Binding Thermodynamics and Kinetics of Both Small Molecules and Flexible Peptides

Binding thermodynamics and kinetics play critical roles in drug design. However, it has proven challenging to efficiently predict ligand binding thermodynamics and kinetics of small molecules and flexible peptides using conventional Molecular Dynamics (cMD), due to limited simulation timescales. Based on our previously developed Ligand Gaussian accelerated Molecular Dynamics (LiGaMD) method, we present a new approach, termed "LiGaMD3", in which we introduce triple boosts into three individual energy terms that play important roles in small-molecule/peptide dissociation, rebinding and system conformational changes to improve the sampling efficiency of small-molecule/peptide interactions with target proteins. To validate the performance of LiGaMD3, MDM2 bound by a small molecule (Nutlin 3) and two highly flexible peptides (PMI and P53) were chosen as model systems. LiGaMD3 could efficiently capture repetitive small-molecule/peptide dissociation and binding events within 2 microsecond simulations. The predicted binding kinetic constant rates and free energies from LiGaMD3 agreed with available experimental values and previous simulation results. Therefore, LiGaMD3 provides a more general and efficient approach to capture dissociation and binding of both small-molecule ligand and flexible peptides, allowing for accurate prediction of their binding thermodynamics and kinetics.

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.

PointDE: Protein Docking Evaluation Using 3D Point Cloud Neural Network

Protein-protein interactions (PPIs) play essential roles in many vital movements and the determination of protein complex structure is helpful to discover the mechanism of PPI. Protein-protein docking is being developed to model the structure of the protein. However, there is still a challenge to selecting the near-native decoys generated by protein-protein docking. Here, we propose a docking evaluation method using 3D point cloud neural network named PointDE. PointDE transforms protein structure to the point cloud. Using the state-of-the-art point cloud network architecture and a novel grouping mechanism, PointDE can capture the geometries of the point cloud and learn the interaction information from the protein interface. On public datasets, PointDE surpasses the state-of-the-art method using deep learning. To further explore the ability of our method in different types of protein structures, we developed a new dataset generated by high-quality antibody-antigen complexes. The result in this antibody-antigen dataset shows the strong performance of PointDE, which will be helpful for the understanding of PPI mechanisms.

Ligand Gaussian Accelerated Molecular Dynamics 2 (LiGaMD2): Improved Calculations of Ligand Binding Thermodynamics and Kinetics with Closed Protein Pocket

Ligand binding thermodynamics and kinetics are critical parameters for drug design. However, it has proven challenging to efficiently predict ligand binding thermodynamics and kinetics from molecular simulations due to limited simulation timescales. Protein dynamics, especially in the ligand binding pocket, often plays an important role in ligand binding. Based on our previously developed Ligand Gaussian accelerated molecular dynamics (LiGaMD), here we present LiGaMD2 in which a selective boost potential was applied to both the ligand and protein residues in the binding pocket to improve sampling of ligand binding and dissociation. To validate the performance of LiGaMD2, the T4 lysozyme (T4L) mutants with open and closed pockets bound by different ligands were chosen as model systems. LiGaMD2 could efficiently capture repetitive ligand dissociation and binding within microsecond simulations of all T4L systems. The obtained ligand binding kinetic rates and free energies agreed well with available experimental values and previous modeling results. Therefore, LiGaMD2 provides an improved approach to sample opening of closed protein pockets for ligand dissociation and binding, thereby allowing for efficient calculations of ligand binding thermodynamics and kinetics.

Using Selectively Scaled Molecular Dynamics Simulations to Assess Ligand Poses in RNA Aptamers

Predicting the structure (or pose) of RNA-ligand complexes is an important problem in RNA structural biology. Although one could use computational docking to rapidly sample putative poses of RNA-ligand complexes, accurately discriminating the native-like poses from non-native, decoy poses remains a formidable challenge. Here, we started from the assumption that native-like RNA-ligand poses are less likely to dissociate during molecular dynamics simulations, and then we used enhanced simulations to promote ligand unbinding for diverse poses of a handful of RNA aptamer-ligand complexes. By fitting unbinding profiles obtained from the simulations to a single exponential, we identified the characteristic decay time (τ) as particularly effective at resolving native poses from decoys. We also found that a simple regression model trained to predict the simulation-derived parameters directly from structure could also discriminate ligand poses for similar RNA aptamers. Characterizing the unbinding properties of individual poses may thus be an effective strategy for enhancing pose prediction for ligands interacting with RNA aptamers. A similar strategy might be applicable to other ligandable RNAs; however, further analysis will be required to confirm this hypothesis.

Using steered molecular dynamic tension for assessing quality of computational protein structure models

The native structures of proteins, except for notable exceptions of intrinsically disordered proteins, in general take their most stable conformation in the physiological condition to maintain their structural framework so that their biological function can be properly carried out. Experimentally, the stability of a protein can be measured by several means, among which the pulling experiment using the atomic force microscope (AFM) stands as a unique method. AFM directly measures the resistance from unfolding, which can be quantified from the observed force-extension profile. It has been shown that key features observed in an AFM pulling experiment can be well reproduced by computational molecular dynamics simulations. Here, we applied computational pulling for estimating the accuracy of computational protein structure models under the hypothesis that the structural stability would positively correlated with the accuracy, i.e. the closeness to the native, of a model. We used in total 4929 structure models for 24 target proteins from the Critical Assessment of Techniques of Structure Prediction (CASP) and investigated if the magnitude of the break force, that is, the force required to rearrange the model's structure, from the force profile was sufficient information for selecting near-native models. We found that near-native models can be successfully selected by examining their break forces suggesting that high break force indeed indicates high stability of models. On the other hand, there were also near-native models that had relatively low peak forces. The mechanisms of the stability exhibited by the break forces were explored and discussed.

Monoclonal Antibody Discovery Based on Precise Selection of Single Transgenic Hybridomas with an On-Cell-Surface and Antigen-Specific Anchor

Hybridoma technology is widely used for monoclonal antibody (mAb) discovery, whereas the generation and identification of single hybridomas by the limiting dilution method (LDM) are tedious, inefficient, and time- and cost-consuming, especially for hapten molecules. Here, we describe a single transgenic hybridoma selection method (STHSM) that employs a transgenic Sp2/0 with an artificial and stable on-cell-surface anchor. The anchor was designed by combining the truncated variant transmembrane domain of EGFR with a biotin acceptor peptide AVI-tag, which was stably integrated into the genome of Sp2/0 via a piggyBac transposon. To ensure the subsequent precise selection of the hybridoma, the number of on-cell-surface anchors of the transfected Sp2/0 for fusion with immunized splenocytes was further normalized by flow cytometry at the single cell level. Then the single antigen-specific transgenic hybridomas were precisely identified and automatically selected using a CellenONE platform based on the fluorescence assay of the on-cell-surface anchor with the corresponding secreted antigen-specific mAb. The STHSM produced 579 single chloramphenicol (CAP)-specific transgenic hybridomas with a positive rate of 62.7% in 10 plates within 2 h by one-step selection, while only 12 single CAP-specific hybridomas with a positive rate of 6.3% in 40 plates required at least 32 days using the LDM with multiple subcloning steps. The best affinity of mAbs from the STHSM was more than 2-fold higher than that of those from the LDM, and this was mainly due to the preaffinity selection based on the on-cell-surface anchors and more interactions between the mAb and CAP. Then the mAbs from the STHSM and LDM were used to develop an immunoassay for CAP in spiked and natural biological samples. The method displayed satisfactory sensitivity, accuracy, and precision, demonstrating that the STHSM we developed is a versatile, practical, and efficient method for mAb discovery.

Protein-Protein Interaction-Gaussian Accelerated Molecular Dynamics (PPI-GaMD): Characterization of Protein Binding Thermodynamics and Kinetics

Protein-protein interactions (PPIs) play key roles in many fundamental biological processes such as cellular signaling and immune responses. However, it has proven challenging to simulate repetitive protein association and dissociation in order to calculate binding free energies and kinetics of PPIs due to long biological timescales and complex protein dynamics. To address this challenge, we have developed a new computational approach to all-atom simulations of PPIs based on a robust Gaussian accelerated molecular dynamics (GaMD) technique. The method, termed "PPI-GaMD", selectively boosts interaction potential energy between protein partners to facilitate their slow dissociation. Meanwhile, another boost potential is applied to the remaining potential energy of the entire system to effectively model the protein's flexibility and rebinding. PPI-GaMD has been demonstrated on a model system of the ribonuclease barnase interactions with its inhibitor barstar. Six independent 2 μs PPI-GaMD simulations have captured repetitive barstar dissociation and rebinding events, which enable calculations of the protein binding thermodynamics and kinetics simultaneously. The calculated binding free energies and kinetic rate constants agree well with the experimental data. Furthermore, PPI-GaMD simulations have provided mechanistic insights into barstar binding to barnase, which involves long-range electrostatic interactions and multiple binding pathways, being consistent with previous experimental and computational findings of this model system. In summary, PPI-GaMD provides a highly efficient and easy-to-use approach for binding free energy and kinetics calculations of PPIs.

Challenges and frontiers of computational modelling of biomolecular recognition

Biomolecular recognition including binding of small molecules, peptides and proteins to their target receptors plays a key role in cellular function and has been targeted for therapeutic drug design. However, the high flexibility of biomolecules and slow binding and dissociation processes have presented challenges for computational modelling. Here, we review the challenges and computational approaches developed to characterise biomolecular binding, including molecular docking, molecular dynamics simulations (especially enhanced sampling) and machine learning. Further improvements are still needed in order to accurately and efficiently characterise binding structures, mechanisms, thermodynamics and kinetics of biomolecules in the future.

NbX: Machine Learning-Guided Re-Ranking of Nanobody-Antigen Binding Poses

Modeling the binding pose of an antibody is a prerequisite to structure-based affinity maturation and design. Without knowing a reliable binding pose, the subsequent structural simulation is largely futile. In this study, we have developed a method of machine learning-guided re-ranking of antigen binding poses of nanobodies, the single-domain antibody which has drawn much interest recently in antibody drug development. We performed a large-scale self-docking experiment of nanobody–antigen complexes. By training a decision tree classifier through mapping a feature set consisting of energy, contact and interface property descriptors to a measure of their docking quality of the refined poses, significant improvement in the median ranking of native-like nanobody poses by was achieved eightfold compared with ClusPro and an established deep 3D CNN classifier of native protein–protein interaction. We further interpreted our model by identifying features that showed relatively important contributions to the prediction performance. This study demonstrated a useful method in improving our current ability in pose prediction of nanobodies.

Native or Non-Native Protein-Protein Docking Models? Molecular Dynamics to the Rescue

Molecular docking excels at creating a plethora of potential models of protein-protein complexes. To correctly distinguish the favorable, native-like models from the remaining ones remains, however, a challenge. We assessed here if a protocol based on molecular dynamics (MD) simulations would allow distinguishing native from non-native models to complement scoring functions used in docking. To this end, the first models for 25 protein-protein complexes were generated using HADDOCK. Next, MD simulations complemented with machine learning were used to discriminate between native and non-native complexes based on a combination of metrics reporting on the stability of the initial models. Native models showed higher stability in almost all measured properties, including the key ones used for scoring in the Critical Assessment of PRedicted Interaction (CAPRI) competition, namely the positional root mean square deviations and fraction of native contacts from the initial docked model. A random forest classifier was trained, reaching a 0.85 accuracy in correctly distinguishing native from non-native complexes. Reasonably modest simulation lengths of the order of 50-100 ns are sufficient to reach this accuracy, which makes this approach applicable in practice.

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.

Thermodynamic Integration in 3n Dimensions without Biases or Alchemy for Protein Interactions

Thermodynamic integration (TI), a powerful formalism for computing Gibbs free energy, has been implemented for many biophysical processes with alchemical schemes that require delicate human efforts to choose/design biasing potentials for sampling the desired biophysical events and to remove their artifactitious consequences afterwards. Theoretically, an alchemical scheme is exact but practically, an unsophisticated implementation of this exact formula can cause error amplifications. Small relative errors in the input parameters can be amplified many times in their propagation into the computed free energy [due to subtraction of similar numbers such as (105 ± 5)‒(100 ± 5) = 5 ± 7]. In this paper, we present an unsophisticated implementation of TI in 3n dimensions (3nD) (n=1,2,3…) for the potential of mean force along a 3nD path connecting one state in the bound state ensemble to one state in the unbound state ensemble. Fluctuations in these 3nD are integrated in the bound and unbound state ensembles but not along the 3nD path. Using TI3nD, we computed the standard binding free energies of three protein complexes: trometamol in Salmonella effector SpvD (n=1), biotin-avidin (n=2), and Colicin E9 endonuclease with cognate immunity protein Im9 (n=3). We employed three different protocols in three independent computations of E9-Im9 to show TI3nD's robustness. We also computed the hydration energies of ten biologically relevant compounds (n=1 for water, acetamide, urea, glycerol, trometamol, ammonium and n=2 for erythritol, 1,3-propanediol, xylitol, biotin). Each of the 15 computations is accomplishable within one (for hydration) to ten (for E9-Im9) days on an inexpensive GPU workstation. The computed results all agree with the available experimental data.

Molecular Dynamics Simulation of Transmembrane Transport of Chloride Ions in Mutants of Channelrhodopsin

Channelrhodopsins (ChRs) are light-gated transmembrane cation channels which are widely used for optogenetic technology. Replacing glutamate located at the central gate of the ion channel with positively charged amino acid residues will reverse ion selectivity and allow anion conduction. The structures and properties of the ion channel, the transport of chloride, and potential of mean force (PMF) of the chimera protein (C1C2) and its mutants, EK-TC, ER-TC and iChloC, were investigated by molecular dynamics simulation. The results show that the five-fold mutation in E122Q-E129R-E140S-D195N-T198C (iChloC) increases the flexibility of the transmembrane channel protein better than the double mutations in EK-TC and ER-TC, and results in an expanded ion channel pore size and decreased steric resistance. The iChloC mutant was also found to have a higher affinity for chloride ions and, based on surface electrostatic potential analysis, provides a favorable electrostatic environment for anion conduction. The PMF free energy curves revealed that high affinity Cl^- binding sites are generated near the central gate of the three mutant proteins. The energy barriers for the EK-TC and ER-TC were found to be much higher than that of iChloC. The results suggest that the transmembrane ion channel of iChloC protein is better at facilitating the capture and transport of chloride ions.

Formation Mechanism of Ion Channel in Channelrhodopsin-2: Molecular Dynamics Simulation and Steering Molecular Dynamics Simulations

Channelrhodopsin-2 (ChR2) is a light-activated and non-selective cationic channel protein that can be easily expressed in specific neurons to control neuronal activity by light. Although ChR2 has been extensively used as an optogenetic tool in neuroscience research, the molecular mechanism of cation channel formation following retinal photoisomerization in ChR2 is not well understood. In this paper, studies of the closed and opened state ChR2 structures are presented. The formation of the cationic channel is elucidated in atomic detail using molecular dynamics simulations on the all-trans-retinal (ChR2-trans) configuration of ChR2 and its isomerization products, 13-cis-retinal (ChR2-cis) configuration, respectively. Photoisomerization of the retinal-chromophore causes the destruction of interactions among the crucial residues (e.g., E90, E82, N258, and R268) around the channel and the extended H-bond network mediated by numerous water molecules, which opens the pore. Steering molecular dynamics (SMD) simulations show that the electrostatic interactions at the binding sites in intracellular gate (ICG) and central gate (CG) can influence the transmembrane transport of Na⁺ in ChR2-cis obviously. Potential of mean force (PMF) constructed by SMD and umbrella sampling also found the existing energy wells at these two binding sites during the transportation of Na⁺. These wells partly hinder the penetration of Na⁺ into cytoplasm through the ion channel. This investigation provides a theoretical insight on the formation mechanism of ion channels and the mechanism of ion permeation.

The design of novel inhibitors for treating cancer by targeting CDC25B through disruption of CDC25B-CDK2/Cyclin A interaction using computational approaches

Cell division cycle 25B is a key cell cycle regulator and widely considered as potent clinical drug target for cancers. This research focused on identifying potential compounds in theory which are able to disrupt transient interactions between CDC25B and its CDK2/Cyclin A substrate.By using the method of ZDOCK and RDOCK, the most optimized 3D structure of CDK2/Cyclin A in complex with CDC25B was constructed and validated using two methods: 1) the superimposition of proteins; 2) analysis of the hydrogen bond distances of Arg 488(N1)-Asp 206(OD1), Arg 492(NE)-Asp 206(OD1), Arg 492(N1)-Asp 206(OD2) and Tyr 497(NE)-Asp 210(OD1). A series of new compounds was gained through searching the fragment database derived from ZINC based on the known inhibitor-compound 7 by the means of "replace fragment" technique. The compounds acquired via meeting the requirements of the absorption, distribution, metabolism, and excretion (ADME) predictions. Finally, 12 compounds with better binding affinity were identified. The comp#1, as a representative, was selected to be synthesized and assayed for their CDC25B inhibitory activities. The comp#1 exhibited mild inhibitory activities against human CDC25B with IC50 values at about 39.02 μM. Molecular Dynamic (MD) simulation revealed that the new inhibitor-comp#1 had favorable conformations for binding to CDC25B and disturbing the interactions between CDC25B and CDK2/Cyclin A.

Computing the binding affinity of a ligand buried deep inside a protein with the hybrid steered molecular dynamics

Computing the ligand-protein binding affinity (or the Gibbs free energy) with chemical accuracy has long been a challenge for which many methods/approaches have been developed and refined with various successful applications. False positives and, even more harmful, false negatives have been and still are a common occurrence in practical applications. Inevitable in all approaches are the errors in the force field parameters we obtain from quantum mechanical computation and/or empirical fittings for the intra- and inter-molecular interactions. These errors propagate to the final results of the computed binding affinities even if we were able to perfectly implement the statistical mechanics of all the processes relevant to a given problem. And they are actually amplified to various degrees even in the mature, sophisticated computational approaches. In particular, the free energy perturbation (alchemical) approaches amplify the errors in the force field parameters because they rely on extracting the small differences between similarly large numbers. In this paper, we develop a hybrid steered molecular dynamics (hSMD) approach to the difficult binding problems of a ligand buried deep inside a protein. Sampling the transition along a physical (not alchemical) dissociation path of opening up the binding cavity---pulling out the ligand---closing back the cavity, we can avoid the problem of error amplifications by not relying on small differences between similar numbers. We tested this new form of hSMD on retinol inside cellular retinol-binding protein 1 and three cases of a ligand (a benzylacetate, a 2-nitrothiophene, and a benzene) inside a T4 lysozyme L99A/M102Q(H) double mutant. In all cases, we obtained binding free energies in close agreement with the experimentally measured values. This indicates that the force field parameters we employed are accurate and that hSMD (a brute force, unsophisticated approach) is free from the problem of error amplification suffered by many sophisticated approaches in the literature.