Clustering Protein Binding Pockets and Identifying Potential Drug Interactions: A Novel Ligand-Based Featurization Method.

Protein-ligand interactions are essential to drug discovery and drug development efforts. Desirable on-target or multitarget interactions are the first step in finding an effective therapeutic, while undesirable off-target interactions are the first step in assessing safety. In this work, we introduce a novel ligand-based featurization and mapping of human protein pockets to identify closely related protein targets and to project novel drugs into a hybrid protein-ligand feature space to identify their likely protein interactions. Using structure-based template matches from PDB, protein pockets are featured by the ligands that bind to their best co-complex template matches. The simplicity and interpretability of this approach provide a granular characterization of the human proteome at the protein-pocket level instead of the traditional protein-level characterization by family, function, or pathway. We demonstrate the power of this featurization method by clustering a subset of the human proteome and evaluating the predicted cluster associations of over 7000 compounds.

Improved Protein-Ligand Binding Affinity Prediction with Structure-Based Deep Fusion Inference.

Predicting accurate protein-ligand binding affinities is an important task in drug discovery but remains a challenge even with computationally expensive biophysics-based energy scoring methods and state-of-the-art deep learning approaches. Despite the recent advances in the application of deep convolutional and graph neural network-based approaches, it remains unclear what the relative advantages of each approach are and how they compare with physics-based methodologies that have found more mainstream success in virtual screening pipelines. We present fusion models that combine features and inference from complementary representations to improve binding affinity prediction. This, to our knowledge, is the first comprehensive study that uses a common series of evaluations to directly compare the performance of three-dimensional (3D)-convolutional neural networks (3D-CNNs), spatial graph neural networks (SG-CNNs), and their fusion. We use temporal and structure-based splits to assess performance on novel protein targets. To test the practical applicability of our models, we examine their performance in cases that assume that the crystal structure is not available. In these cases, binding free energies are predicted using docking pose coordinates as the inputs to each model. In addition, we compare these deep learning approaches to predictions based on docking scores and molecular mechanic/generalized Born surface area (MM/GBSA) calculations. Our results show that the fusion models make more accurate predictions than their constituent neural network models as well as docking scoring and MM/GBSA rescoring, with the benefit of greater computational efficiency than the MM/GBSA method. Finally, we provide the code to reproduce our results and the parameter files of the trained models used in this work. The software is available as open source at https://github.com/llnl/fast. Model parameter files are available at ftp://gdo-bioinformatics.ucllnl.org/fast/pdbbind2016_model_checkpoints/.

Efficient Computational Modeling of Human Ventricular Activation and Its Electrocardiographic Representation: A Sensitivity Study.

Patient-specific models of the ventricular myocardium, combined with the computational power to run rapid simulations, are approaching the level where they could be used for personalized cardiovascular medicine. A major remaining challenge is determining model parameters from available patient data, especially for models of the Purkinje-myocardial junctions (PMJs): the sites of initial ventricular electrical activation. There are no non-invasive methods for localizing PMJs in patients, and the relationship between the standard clinical ECG and PMJ model parameters is underexplored. Thus, this study aimed to determine the sensitivity of the QRS complex of the ECG to the anatomical location and regional number of PMJs. The QRS complex was simulated using an image-based human torso and biventricular model, and cardiac electrophysiology was simulated using Cardioid. The PMJs were modeled as discrete current injection stimuli, and the location and number of stimuli were varied within initial activation regions based on published experiments. Results indicate that the QRS complex features were most sensitive to the presence or absence of four "seed" stimuli, and adjusting locations of nearby "regional" stimuli provided finer tuning. Decreasing number of regional stimuli by an order of magnitude resulted in virtually no change in the QRS complex. Thus, a minimal 12-stimuli configuration was identified that resulted in physiological excitation, defined by QRS complex feature metrics and ventricular excitation pattern. Overall, the sensitivity results suggest that parameterizing PMJ location, rather than number, be given significantly higher priority in future studies creating personalized ventricular models from patient-derived ECGs.

Adverse drug reaction prediction using scores produced by large-scale drug-protein target docking on high-performance computing machines.

Late-stage or post-market identification of adverse drug reactions (ADRs) is a significant public health issue and a source of major economic liability for drug development. Thus, reliable in silico screening of drug candidates for possible ADRs would be advantageous. In this work, we introduce a computational approach that predicts ADRs by combining the results of molecular docking and leverages known ADR information from DrugBank and SIDER. We employed a recently parallelized version of AutoDock Vina (VinaLC) to dock 906 small molecule drugs to a virtual panel of 409 DrugBank protein targets. L1-regularized logistic regression models were trained on the resulting docking scores of a 560 compound subset from the initial 906 compounds to predict 85 side effects, grouped into 10 ADR phenotype groups. Only 21% (87 out of 409) of the drug-protein binding features involve known targets of the drug subset, providing a significant probe of off-target effects. As a control, associations of this drug subset with the 555 annotated targets of these compounds, as reported in DrugBank, were used as features to train a separate group of models. The Vina off-target models and the DrugBank on-target models yielded comparable median area-under-the-receiver-operating-characteristic-curves (AUCs) during 10-fold cross-validation (0.60-0.69 and 0.61-0.74, respectively). Evidence was found in the PubMed literature to support several putative ADR-protein associations identified by our analysis. Among them, several associations between neoplasm-related ADRs and known tumor suppressor and tumor invasiveness marker proteins were found. A dual role for interstitial collagenase in both neoplasms and aneurysm formation was also identified. These associations all involve off-target proteins and could not have been found using available drug/on-target interaction data. This study illustrates a path forward to comprehensive ADR virtual screening that can potentially scale with increasing number of CPUs to tens of thousands of protein targets and millions of potential drug candidates.

A method to predict blood-brain barrier permeability of drug-like compounds using molecular dynamics simulations.

The blood-brain barrier (BBB) is formed by specialized tight junctions between endothelial cells that line brain capillaries to create a highly selective barrier between the brain and the rest of the body. A major problem to overcome in drug design is the ability of the compound in question to cross the BBB. Neuroactive drugs are required to cross the BBB to function. Conversely, drugs that target other parts of the body ideally should not cross the BBB to avoid possible psychotropic side effects. Thus, the task of predicting the BBB permeability of new compounds is of great importance. Two gold-standard experimental measures of BBB permeability are logBB (the concentration of drug in the brain divided by concentration in the blood) and logPS (permeability surface-area product). Both methods are time-consuming and expensive, and although logPS is considered the more informative measure, it is lower throughput and more resource intensive. With continual increases in computer power and improvements in molecular simulations, in silico methods may provide viable alternatives. Computational predictions of these two parameters for a sample of 12 small molecule compounds were performed. The potential of mean force for each compound through a 1,2-dioleoyl-sn-glycero-3-phosphocholine bilayer is determined by molecular dynamics simulations. This system setup is often used as a simple BBB mimetic. Additionally, one-dimensional position-dependent diffusion coefficients are calculated from the molecular dynamics trajectories. The diffusion coefficient is combined with the free energy landscape to calculate the effective permeability (Peff) for each sample compound. The relative values of these permeabilities are compared to experimentally determined logBB and logPS values. Our computational predictions correlate remarkably well with both logBB (R(2) = 0.94) and logPS (R(2) = 0.90). Thus, we have demonstrated that this approach may have the potential to provide reliable, quantitatively predictive BBB permeability, using a relatively quick, inexpensive method.

Understanding a substrate's product regioselectivity in a family of enzymes: a case study of acetaminophen binding in cytochrome P450s.

Product regioselectivity as influenced by molecular recognition is a key aspect of enzyme catalysis. We applied large-scale two-dimensional (2D) umbrella sampling (USP) simulations to characterize acetaminophen (APAP) binding in the active sites of the family of Cytochrome P450 (CYP) enzymes as a case study to show the different regioselectivity exhibited by a single substrate in comparative enzymes. Our results successfully explain the experimentally observed product regioselectivity for all five human CYPs included in this study, demonstrating that binding events play an important role in determining regioselectivity. In CYP2C9 and CYP3A4, weak interactions in an overall large active site cavity result in a fairly small binding free energy difference between APAP reactive binding states, consistent with experimental results that show little preference for resulting metabolites. In contrast, in CYP1A2 and CYP2E1, APAP is strongly restrained by a compact binding pocket, leading to a preferred binding conformation. The calculated binding equilibrium of APAP within the compact active site of CYP2A6 is able to predict the experimentally documented product ratios and is also applied to explain APAP regioselectivity in CYP1A2 and CYP2C9. APAP regioselectivity seems to be related to the selectivity for one binding conformation over another binding conformation as dictated by the size and shape of the active site. Additionally, unlike docking and molecular dynamics (MD), our free energy calculations successfully reproduced a unique APAP pose in CYP3A4 that had been reported experimentally, suggesting this approach is well suited to find the realistic binding pose and the lowest-energy starting structure for studying the chemical reaction step in the future.

Developing an approach for first-principles catalyst design: application to carbon-capture catalysis.

An approach to catalyst design is presented in which local potential energy surface models are first built to elucidate design principles and then used to identify larger scaffold motifs that match the target geometries. Carbon sequestration via hydration is used as the model reaction, and three- and four-coordinate sp(2) or sp(3) nitrogen-ligand motifs are considered for Zn(II) metals. The comparison of binding, activation and product release energies over a large range of interaction distances and angles suggests that four-coordinate short Zn(II)-Nsp(3) bond distances favor a rapid turnover for CO2 hydration. This design strategy is then confirmed by computationally characterizing the reactivity of a known mimic over a range of metal-nitrogen bond lengths. A search of existing catalysts in a chemical database reveals structures that match the target geometry from model calculations, and subsequent calculations have identified these structures as potentially effective for CO2 hydration and sequestration.

Toward fully automated high performance computing drug discovery: a massively parallel virtual screening pipeline for docking and molecular mechanics/generalized Born surface area rescoring to improve enrichment.

In this work we announce and evaluate a high throughput virtual screening pipeline for in-silico screening of virtual compound databases using high performance computing (HPC). Notable features of this pipeline are an automated receptor preparation scheme with unsupervised binding site identification. The pipeline includes receptor/target preparation, ligand preparation, VinaLC docking calculation, and molecular mechanics/generalized Born surface area (MM/GBSA) rescoring using the GB model by Onufriev and co-workers [J. Chem. Theory Comput. 2007, 3, 156-169]. Furthermore, we leverage HPC resources to perform an unprecedented, comprehensive evaluation of MM/GBSA rescoring when applied to the DUD-E data set (Directory of Useful Decoys: Enhanced), in which we selected 38 protein targets and a total of ∼0.7 million actives and decoys. The computer wall time for virtual screening has been reduced drastically on HPC machines, which increases the feasibility of extremely large ligand database screening with more accurate methods. HPC resources allowed us to rescore 20 poses per compound and evaluate the optimal number of poses to rescore. We find that keeping 5-10 poses is a good compromise between accuracy and computational expense. Overall the results demonstrate that MM/GBSA rescoring has higher average receiver operating characteristic (ROC) area under curve (AUC) values and consistently better early recovery of actives than Vina docking alone. Specifically, the enrichment performance is target-dependent. MM/GBSA rescoring significantly out performs Vina docking for the folate enzymes, kinases, and several other enzymes. The more accurate energy function and solvation terms of the MM/GBSA method allow MM/GBSA to achieve better enrichment, but the rescoring is still limited by the docking method to generate the poses with the correct binding modes.

Molecular recognition of aromatic rings by flavin: electrostatics and dispersion determine ring positioning above isoalloxazine.

Aromatic stacking interactions between isoalloxazine (ISA) of flavin and three prototypical aromatics (benzene, pyridine, chlorobenzene) were investigated using electronic structure calculations with Monte Carlo simulated annealing. The Effective Fragment Potential (EFP) method was used to locate the low-energy equilibrium configurations for the three dimer systems. These structures were further characterized through DFT (M06-2X) and MP2 calculations. One equilibrium configuration exists for ISA-benzene; characterizing the stacked dimer surface revealed a steep, single-welled potential that funnels benzene directly between rings II and III, positioning a substituent hydrogen adjacent to the redox-active N5. ISA-pyridine and ISA-chlorobenzene minimum-energy structures contain the aromatic ring in very similar position to that in ISA-benzene. However, the added rotational degree of freedom leads to two distinct binding motifs, having approximately antiparallel or parallel dipole moment alignment with ISA. The existence of the latter binding configuration was unexpected but is explained by the shape of the ISA electrostatic potential. Dispersion is the primary noncovalent interaction driving the positioning of aromatic rings above ISA, while electrostatics determine the orientation in dipole-containing substituted benzenes. The interplay of these interactions can be used to tune molecular recognition properties of synthetic redox cofactors, including positioning desired functional groups adjacent to the redox-active N5.

Comparison and analysis of zinc and cobalt-based systems as catalytic entities for the hydration of carbon dioxide.

In nature, the zinc metalloenzyme carbonic anhydrase II (CAII) efficiently catalyzes the conversion of carbon dioxide (CO2) to bicarbonate under physiological conditions. Many research efforts have been directed towards the development of small molecule mimetics that can facilitate this process and thus have a beneficial environmental impact, but these efforts have met very limited success. Herein, we undertook quantum mechanical calculations of four mimetics, 1,5,9-triazacyclododedacane, 1,4,7,10-tetraazacyclododedacane, tris(4,5-dimethyl-2-imidazolyl)phosphine, and tris(2-benzimidazolylmethyl)amine, in their complexed form either with the Zn(2+) or the Co(2+) ion and studied their reaction coordinate for CO2 hydration. These calculations demonstrated that the ability of the complex to maintain a tetrahedral geometry and bind bicarbonate in a unidentate manner were vital for the hydration reaction to proceed favorably. Furthermore, these calculations show that the catalytic activity of the examined zinc complexes was insensitive to coordination states for zinc, while coordination states above four were found to have an unfavorable effect on product release for the cobalt counterparts.

Rapid catalytic template searching as an enzyme function prediction procedure.

We present an enzyme protein function identification algorithm, Catalytic Site Identification (CatSId), based on identification of catalytic residues. The method is optimized for highly accurate template identification across a diverse template library and is also very efficient in regards to time and scalability of comparisons. The algorithm matches three-dimensional residue arrangements in a query protein to a library of manually annotated, catalytic residues--The Catalytic Site Atlas (CSA). Two main processes are involved. The first process is a rapid protein-to-template matching algorithm that scales quadratically with target protein size and linearly with template size. The second process incorporates a number of physical descriptors, including binding site predictions, in a logistic scoring procedure to re-score matches found in Process 1. This approach shows very good performance overall, with a Receiver-Operator-Characteristic Area Under Curve (AUC) of 0.971 for the training set evaluated. The procedure is able to process cofactors, ions, nonstandard residues, and point substitutions for residues and ions in a robust and integrated fashion. Sites with only two critical (catalytic) residues are challenging cases, resulting in AUCs of 0.9411 and 0.5413 for the training and test sets, respectively. The remaining sites show excellent performance with AUCs greater than 0.90 for both the training and test data on templates of size greater than two critical (catalytic) residues. The procedure has considerable promise for larger scale searches.

Pyrrolopyrimidine inhibitors of DNA gyrase B (GyrB) and topoisomerase IV (ParE). Part I: Structure guided discovery and optimization of dual targeting agents with potent, broad-spectrum enzymatic activity.

The bacterial topoisomerases DNA gyrase (GyrB) and topoisomerase IV (ParE) are essential enzymes that control the topological state of DNA during replication. The high degree of conservation in the ATP-binding pockets of these enzymes make them appealing targets for broad-spectrum inhibitor development. A pyrrolopyrimidine scaffold was identified from a pharmacophore-based fragment screen with optimization potential. Structural characterization of inhibitor complexes conducted using selected GyrB/ParE orthologs aided in the identification of important steric, dynamic and compositional differences in the ATP-binding pockets of the targets, enabling the design of highly potent pyrrolopyrimidine inhibitors with broad enzymatic spectrum and dual targeting activity.

Approaches to efficiently estimate solvation and explicit water energetics in ligand binding: the use of WaterMap.

INTRODUCTION: Water displacement plays critical role in several phases of drug discovery. Proper treatment of displacing water could improve enrichment in virtual screening and could lead to more successes in lead optimization. WaterMap has recently emerged as a promising approach in this regard; recent implementations of this protocol successfully explained various binding activity that were poorly understood previously, including the well-known super affinity associated with biotin binding to streptavidin. AREAS COVERED: The review briefly discusses implicit and explicit solvent models and focuses on an application of inhomogeneous solvation theory - WaterMap. Furthermore, the review discusses various successful cases where the use of WaterMap explained selectivity in protein-ligand binding and provides discussion of the fundamentals and recently successful implementations of WaterMap. The authors also discuss the limitations of this protocol and list a few approaches that could extend its implementation to more cases. EXPERT OPINION: WaterMap is a powerful tool for calculating the cost of desolvation for structural waters. In some cases, it proved useful in predicting relative binding free energy differences for congeneric ligands. The practical utility of WaterMap hinges in adequate application of the results in the context of all the thermodynamic contributions to binding. Potential improvements as well as integration into methods such like MM-GB/SA could extend its success.

Pyrrolopyrimidine inhibitors of DNA gyrase B (GyrB) and topoisomerase IV (ParE), Part II: development of inhibitors with broad spectrum, Gram-negative antibacterial activity.

The structurally related bacterial topoisomerases DNA gyrase (GyrB) and topoisomerase IV (ParE) have long been recognized as prime candidates for the development of broad spectrum antibacterial agents. However, GyrB/ParE targeting antibacterials with spectrum that encompasses robust Gram-negative pathogens have not yet been reported. Using structure-based inhibitor design, we optimized a novel pyrrolopyrimidine inhibitor series with potent, dual targeting activity against GyrB and ParE. Compounds were discovered with broad antibacterial spectrum, including activity against Pseudomonas aeruginosa, Acinetobacter baumannii and Escherichia coli. Herein we describe the SAR of the pyrrolopyrimidine series as it relates to key structural and electronic features necessary for Gram-negative antibacterial activity.

Message passing interface and multithreading hybrid for parallel molecular docking of large databases on petascale high performance computing machines.

A mixed parallel scheme that combines message passing interface (MPI) and multithreading was implemented in the AutoDock Vina molecular docking program. The resulting program, named VinaLC, was tested on the petascale high performance computing (HPC) machines at Lawrence Livermore National Laboratory. To exploit the typical cluster-type supercomputers, thousands of docking calculations were dispatched by the master process to run simultaneously on thousands of slave processes, where each docking calculation takes one slave process on one node, and within the node each docking calculation runs via multithreading on multiple CPU cores and shared memory. Input and output of the program and the data handling within the program were carefully designed to deal with large databases and ultimately achieve HPC on a large number of CPU cores. Parallel performance analysis of the VinaLC program shows that the code scales up to more than 15K CPUs with a very low overhead cost of 3.94%. One million flexible compound docking calculations took only 1.4 h to finish on about 15K CPUs. The docking accuracy of VinaLC has been validated against the DUD data set by the re-docking of X-ray ligands and an enrichment study, 64.4% of the top scoring poses have RMSD values under 2.0 Å. The program has been demonstrated to have good enrichment performance on 70% of the targets in the DUD data set. An analysis of the enrichment factors calculated at various percentages of the screening database indicates VinaLC has very good early recovery of actives.

Computational Analysis of a Zn-Bound Tris(imidazolyl) Calix[6]arene Aqua Complex: Toward Incorporating Second-Coordination Sphere Effects into Carbonic Anhydrase Biomimetics.

Molecular dynamics simulations and quantum-mechanical calculations were performed to characterize a supramolecular tris(imidazolyl) calix[6]arene Zn(2+) aqua complex, as a biomimetic model for the catalyzed hydration of carbon dioxide to bicarbonate, H2O + CO2 → H(+) + HCO3(-). On the basis of potential-of-mean-force (PMF) calculations, stable conformations had distorted 3-fold symmetry and supported either one or zero encapsulated water molecules. The conformation with an encapsulated water molecule is calculated to be lower in free energy than the conformation with an empty cavity (ΔG = 1.2 kcal/mol) and is the calculated free-energy minimum in solution. CO2 molecule partitioning into the cavity is shown to be very facile, proceeding with a barrier of 1.6 kcal/mol from a weak encounter complex which stabilizes the species by about 1.0 kcal/mol. The stabilization energy of CO2 is calculated to be larger than that of H2O (ΔΔG = 1.4 kcal/mol), suggesting that the complex will preferentially encapsulate CO2 in solution. In contrast, the PMF for a bicarbonate anion entering the cavity is calculated to be repulsive in all nonbonding regions of the cavity, due to the diameter of the calix[6]arene walls. Geometry optimization of the Zn-bound hydroxide complex with an encapsulated CO2 molecule showed that multiple noncovalent interactions direct the reactants into optimal position for nucleophilic addition to occur. The calixarene complex is a structural mimic of the hydrophilic/hydrophobic divide in the enzyme, providing a functional effect for CO2 addition in the catalytic cycle. The results show that Zn-binding calix[6]arene scaffolds can be potential synthetic biomimetics for CO2 hydration catalysis, both in terms of preferentially encapsulating CO2 from solution and by spatially fixing the reactive species inside the cavity.

Tricyclic GyrB/ParE (TriBE) inhibitors: a new class of broad-spectrum dual-targeting antibacterial agents.

Increasing resistance to every major class of antibiotics and a dearth of novel classes of antibacterial agents in development pipelines has created a dwindling reservoir of treatment options for serious bacterial infections. The bacterial type IIA topoisomerases, DNA gyrase and topoisomerase IV, are validated antibacterial drug targets with multiple prospective drug binding sites, including the catalytic site targeted by the fluoroquinolone antibiotics. However, growing resistance to fluoroquinolones, frequently mediated by mutations in the drug-binding site, is increasingly limiting the utility of this antibiotic class, prompting the search for other inhibitor classes that target different sites on the topoisomerase complexes. The highly conserved ATP-binding subunits of DNA gyrase (GyrB) and topoisomerase IV (ParE) have long been recognized as excellent candidates for the development of dual-targeting antibacterial agents with broad-spectrum potential. However, to date, no natural product or small molecule inhibitors targeting these sites have succeeded in the clinic, and no inhibitors of these enzymes have yet been reported with broad-spectrum antibacterial activity encompassing the majority of Gram-negative pathogens. Using structure-based drug design (SBDD), we have created a novel dual-targeting pyrimidoindole inhibitor series with exquisite potency against GyrB and ParE enzymes from a broad range of clinically important pathogens. Inhibitors from this series demonstrate potent, broad-spectrum antibacterial activity against Gram-positive and Gram-negative pathogens of clinical importance, including fluoroquinolone resistant and multidrug resistant strains. Lead compounds have been discovered with clinical potential; they are well tolerated in animals, and efficacious in Gram-negative infection models.

Toward a small molecule, biomimetic carbonic anhydrase model: theoretical and experimental investigations of a panel of zinc(II) aza-macrocyclic catalysts.

A panel of five zinc-chelated aza-macrocycle ligands and their ability to catalyze the hydration of carbon dioxide to bicarbonate, H(2)O + CO(2) → H(+) + HCO(3)(­), was investigated using quantum-mechanical methods and stopped-flow experiments. The key intermediates in the reaction coordinate were optimized using the M06-2X density functional with aug-cc-pVTZ basis set. Activation energies for the first step in the catalytic cycle, nucleophilic CO(2) addition, were calculated from gas-phase optimized transition-state geometries. The computationally derived trend in activation energies was found to not correspond with the experimentally observed rates. However, activation energies for the second, bicarbonate release step, which were estimated using calculated bond dissociation energies, provided good agreement with the observed trend in rate constants. Thus, the joint theoretical and experimental results provide evidence that bicarbonate release, not CO(2) addition, may be the rate-limiting step in CO(2) hydration by zinc complexes of aza-macrocyclic ligands. pH-independent rate constants were found to increase with decreasing Lewis acidity of the ligand-Zn complex, and the trend in rate constants was correlated with molecular properties of the ligands. It is suggested that tuning catalytic efficiency through the first coordination shell of Zn(2+) ligands is predominantly a balance between increasing charge-donating character of the ligand and maintaining the catalytically relevant pK(a) below the operating pH.

Leveraging structural information for the discovery of new drugs: computational methods.

Escalating problems with drug resistance continue to compromise the effectiveness of commercial antibiotics, necessitating the search for novel classes of antimicrobial agents. To circumvent problems with resistance, a multitarget single-pharmacophore approach has been employed to discover inhibitors that possess balanced activity against multiple target enzymes. In this chapter, we examine the application of computational techniques, in particular, structure-based drug design approaches, to design new dual-targeting antibacterial agents against bacterial topoisomerases.

Effects of somatic mutations on CDR loop flexibility during affinity maturation.

Prior studies suggest that antibody affinity maturation is achieved, in part, via prearranging the CDRs for binding. The implication is that the entropy cost of binding is reduced and that this rigidification occurs as a consequence of somatic mutations during maturation. However, how these mutations modulate CDR flexibility is unclear. Here, molecular dynamics simulations captured CDR flexibility differences between four mature antibodies (7G12, AZ28, 28B4, and 48G7) and their germline predecessors. Analysis of their trajectories: (1) rationalized how mutations during affinity maturation restrict CDR motility, (2) captured the equilibrium between bound and unbound conformations for the H3 loop of unliganded 7G12, and (3) predicted a set of new mutations that, according to our simulations, should diminish binding by increasing flexibility.