Do AI Models for Protein Structure Prediction Get Electrostatics Right?

Computational approaches could decrease the need for the laborious high-throughput experimental screening that is often required to improve enzymes by mutagenesis. Here, we report that using multiple short molecular dynamics (MD) simulations makes it possible to accurately model enantioselectivity for large numbers of enzyme-substrate combinations at low computational costs. We chose four different haloalkane dehalogenases as model systems because of the availability of a large set of experimental data on the enantioselective conversion of 45 different substrates. To model the enantioselectivity, we quantified the frequency of occurrence of catalytically productive conformations (near attack conformations) for pairs of enantiomers during MD simulations. We found that the angle of nucleophilic attack that leads to carbon-halogen bond cleavage was a critical variable that limited the occurrence of productive conformations; enantiomers for which this angle reached values close to 180° were preferentially converted. A cluster of 20-40 very short (10 ps) MD simulations allowed adequate conformational sampling and resulted in much better agreement to experimental enantioselectivities than single long MD simulations (22 ns), while the computational costs were 50-100 fold lower. With single long MD simulations, the dynamics of enzyme-substrate complexes remained confined to a conformational subspace that rarely changed significantly, whereas with multiple short MD simulations a larger diversity of conformations of enzyme-substrate complexes was observed.

Membrane Assembly of Simple Helix Homo-Oligomers Studied via Molecular Dynamics Simulations

Conducting-State Properties of the KcsA Potassium Channel from Molecular and Brownian Dynamics Simulations

With the emergence of the new SARS-CoV-2 virus, drug repurposing studies have gained substantial importance. Combined with the efficacy of recent improvements in ligand- and target-based virtual screening approaches, virtual screening has become faster and more productive than ever. In the current study, an FDA library of approved drugs and compounds under clinical investigation were screened for their antiviral activity using the antiviral therapeutic activity binary QSAR model of the MetaCore/MetaDrug platform. Among 6733-compound collection, we found 370 compounds with a normalized therapeutic activity value greater than a cutoff of 0.75. Only these selected compounds were used for molecular docking studies against the SARS-CoV-2 main protease (Mpro). After initial short (10 ns) molecular dynamics (MD) simulations with the top-50 docking scored compounds and following molecular mechanics generalized born surface area (MM/GBSA) calculations, top-10 compounds were subjected to longer (100 ns) MD simulations and end-point MM/GBSA estimations. Our virtual screening protocol yielded Cefuroxime pivoxetil, an ester prodrug of second-generation cephalosporin antibiotic Cefuroxime, as being a considerable molecule for drug repurposing against the SARS-CoV-2 Mpro.

AlphaFold 2 (AF2) has placed Molecular Biology in a new era where we can visualize, analyze and interpret the structures and functions of all proteins solely from their primary sequences. We performed AF2 structure predictions for various protein systems, including globular proteins, a multi-domain protein, an intrinsically disordered protein (IDP), a randomized protein, two larger proteins (> 1000 AA), a heterodimer and a homodimer protein complex. Our results show that along with the three dimensional (3D) structures, AF2 also decodes protein sequences into residue flexibilities via both the predicted local distance difference test (pLDDT) scores of the models, and the predicted aligned error (PAE) maps. We show that PAE maps from AF2 are correlated with the distance variation (DV) matrices from molecular dynamics (MD) simulations, which reveals that the PAE maps can predict the dynamical nature of protein residues. Here, we introduce the AF2-scores, which are simply derived from pLDDT scores and are in the range of [0, 1]. We found that for most protein models, including large proteins and protein complexes, the AF2-scores are highly correlated with the root mean square fluctuations (RMSF) calculated from MD simulations. However, for an IDP and a randomized protein, the AF2-scores do not correlate with the RMSF from MD, especially for the IDP. Our results indicate that the protein structures predicted by AF2 also convey information of the residue flexibility, i.e., protein dynamics.

Viscosity calculation from equilibrium molecular dynamics (MD) simulations relies on the traditional Green-Kubo (GK) framework, which integrates the stress autocorrelation function (SACF) over time. While the formalism is exact in the linear response regime, the traditional approach often suffers from poor convergence and requires extensive phase space sampling, which is computationally demanding for soft matter and polymer systems. In this Letter, we introduce a hybrid Green-Kubo (hGK) framework that alleviates these limitations by partitioning the SACF into two physically meaningful regimes: (i) a short time ballistic component extracted directly from short MD simulations and (ii) a long time relaxation tail represented using analytically motivated functions, ϕ(τ), fitted only to short trajectories such that η = (V/kBT)[1/6∑αβ∫0τl⟨Pαβ(t) Pαβ(t + τ )⟩ dτ + ∫τl∞ϕ(τ) dτ]. This strategy bypasses the need for extensive sampling while preserving the exact GK framework in short times. Benchmarking against SPC/E water confirms excellent agreement with established results, and we further demonstrate the efficacy of the method for challenging electrolyte systems (EC-LiTFSI and PEO-LiTFSI), for which the GK framework fails to converge. The computational savings are substantial, with reductions of several orders of magnitude in required sampling achieved without compromising predictive accuracy. We also discuss the limitations of the hGK framework and outline clear avenues for refinement, including optimal tail selection and robust identification of relaxation regimes in noisy stress data. The hGK framework presented in this Letter provides a conceptually simple, broadly applicable, and computationally efficient route for viscosity prediction in molecular liquids, polymer melts, and ionically conducting soft materials.

Staphylococcus aureus remains a critical ESKAPE pathogen due to its multidrug resistance and reliance on virulence-associated enzymes. In this study, an integrative in silico pipeline combining molecular docking, redocking-based validation, molecular dynamics (MD) simulations, and pharmacokinetic profiling was applied to prioritize Psidium guajava metabolites targeting two functionally distinct proteins: the essential peptidoglycan biosynthesis enzyme MurE and the antivirulence factor sortase A (SrtA). Twenty-one metabolites were docked against both targets, and reference redocking experiments were performed using the co-crystallized ligand for MurE and quercetin for SrtA to validate the docking protocol. Flavonoids quercetin and myricetin exhibited the strongest binding affinities toward MurE (≈-9.8 to -9.9 kcal·mol -1 ) and displayed stable MD behavior over 100 ns, characterized by low backbone RMSD values (0.25-0.40 nm), limited RMSF fluctuations confined to terminal regions, a stable radius of gyration, and consistent potential energy profiles, indicating persistent interactions within the MurE catalytic site. In contrast, SrtA complexes with rutin and oleanolic acid showed pronounced conformational rearrangements during MD simulations, with large backbone RMSD excursions (≈2-9 nm) and fluctuating compactness, with partial stabilization emerging toward the end of the 100 ns trajectory (approximately after ~60-80 ns), consistent with dynamic or transient binding rather than stable inhibition. Pharmacokinetic analyses highlighted quercetin and myricetin as the most drug-like candidates among the evaluated metabolites. Overall, the results support a complementary, target-selective strategy, identifying MurE as a robust enzymatic target for P. guajava flavonoids, while SrtA interactions are better interpreted as potential antivirulence modulation rather than strong dual-target inhibition. This refined framework provides a realistic and mechanistically grounded basis for prioritizing plant-derived scaffolds against S. aureus.

Molecular dynamics (MD) simulations together with accurate physics-based force fields are employed to determine the mobility of atomic oxygen in amorphous ice at low temperatures, characteristic for conditions in interstellar ices. From the simulations it is found that the mobility of atomic oxygen ranges from 60 to 480 angstroms2 ns(-1) in amorphous ice at temperatures between 50 and 200 K. Hence, the simulations establish that atomic oxygen is mobile to a certain degree and a chemical mechanism for water formation involving oxygen mobility is a realistic scenario. This is also confirmed by the computed migration barriers for oxygen diffusion by multiple umbrella sampling simulations, which yield barriers for diffusion in the range of 0.7-1.9 kcal mol(-1). The physics-based force field--based on a multipolar expansion of the electrostatic interactions--yields more pronounced energetics for oxygen migration pathways compared to the conventional point-charge models employed in typical simulations. Once formed, the computed solvation free energy suggests that atomic oxygen thermodynamically prefers to be localized inside amorphous ice and is available for chemical reaction, which may be relevant to water formation in and on grains.

Graphs have been proved to be a useful mathematical representation for a broad variety of real-world complex systems, and the structure prediction on graphs refers to estimating the potential relationship between the objects from the observed structures, being fundamental in many data analysis applications, such as network alignment, network reconstruction, and link prediction. Accordingly, in data publishing, it is necessary to regulate the structural predictability of graphs against inference attack to protect the sensitive information of the data generators. In contrast to the existing works about graph structure perturbation for node ranking, information diffusion, and so on, the structural predictability optimization problem, i.e., reducing the accuracy of sensitive relationships inference in graphs, has not been extensively studied. This paper presents an active learning algorithm that selects the most representative links to be perturbed, thus regulating the structural predictability of graphs, that is, removing as few as possible links to undermine the regularity level of graphs, which forms the foundation of inference attack methods. Specifically, with the assumption that the substructure with higher regularity level contains more regular equivalence components and has more equivalent paths supplied for the random walk processes, random walk-based link importance measuring algorithm is proposed to identify the representative links. The structural regularity metric, measuring the structural predictability of graphs, is also introduced to guide the link perturbation for structural predictability optimization. The extensive experiments on artificial and real-world data sets demonstrate the effectiveness of the proposed structural predictability optimization method. Specifically, the method can learn the role of links accurately in term of graph organization, and the performance of structure inference on graphs can be deteriorated effectively by representative link-based perturbation.

Carboxypeptidase G(2) (CPG(2)) is a zinc-metalloenzyme employed in a range of cancer chemotherapy strategies by activating selectively nontoxic prodrugs into cytotoxic drugs in tumor as well as in the treatment of intoxication caused by high-doses of the anticancer drug methotrexate (MTX). CPG(2) catalyzes the hydrolytic cleavage of C-terminal of glutamate moiety from folic acid and analogues. Regardless of its extensive application, its mechanism of catalysis has not yet been determined and, so far, no co-crystallized complex has been published. So, in this study, molecular docking and a short molecular dynamics (MD) simulation sampling scheme, as a function of temperature, were performed to investigate a possible binding mode for MTX, a recognized substrate of CPG(2). The findings suggested that MTX interacts possibly in quite specific points of the CPG(2) active site, which are probably responsible for the molecular recognition and cleavage procedures. The MTX substrate fits well in the catalytic site by accommodating the pteridine moiety in an adjacent pocket to the active site whereas a glutamate moiety is pointed toward the protein surface. Additionally, a glutamate residue can interact with a crystallization water molecule in the active site, supporting its activation as a nucleophilic group.

The rapid emergence of COVID-19 pandemics has posed humans particularly vulnerable to the novel SARS-CoV-2 virus. Since de novo drug discovery is both expensive and time-consuming, drug repurposing approaches are believed to be of particular help. The SARS-CoV-2 spike (S) protein is known to attach human angiotensin-converting enzyme-2 (hACE2) through its receptor-binding domain (RBD). We screened 1930 FDA-approved ligands for the selection of optimal ones blocking this interaction. Virtual screening predicted top 25 ligands docking to any of the reported binding sites. After exclusion of those ligands which were unsuitable for systemic use, the remaining 69 RBD-ligand complexes were screened based on the masking capacity of the amino acid residues engaged in RBD-hACE2 interaction, excluding 47 RBD-ligand complexes. A short molecular dynamics (MD) simulation analysis identified 11 globally stable complexes with the lowest RMSD (root-mean-square deviation). Next, a moderately long MD analysis revealed those six RBD-ligand complexes with the lowest RMSD variation, as a measure of global stability. Finally, a long MD analysis revealed two select candidate ligands, including ritonavir and naloxegol, highly stabilizing those key residues engaged in RBD-hACE2 interaction. A similar MD analysis of a few antiviral drugs which are under clinical trials or approved for COVID-19 treatment showed them inferior to both select ligands in terms of stabilizing the RBD globally and locally at binding sites. Because of the crucial role of the S protein in virus virulence, our results highly propose ritonavir and naloxegol as the potentially helpful therapeutics against COVID-19, mandating appropriate clinical trials. Communicated by Ramaswamy H. Sarma

The emergence of antimicrobial resistance threatens modern medicine and necessitates more personalized treatment of bacterial infections. Sequencing the whole genome of the pathogen(s) in a clinical sample offers one way to improve clinical microbiology diagnostic services, and has already been adopted for tuberculosis in some countries. A key weakness of a genetics clinical microbiology is it cannot return a result for rare or novel genetic variants and therefore predictive methods are required. Non-synonymous mutations in the S. aureusdfrB gene can be successfully classified as either conferring resistance (or not) by calculating their effect on the binding free energy of the antibiotic, trimethoprim. The underlying approach, alchemical free energy methods, requires large numbers of molecular dynamics simulations to be run. We show that a large number (N = 15) of binding free energies calculated from a series of very short (50 ps) molecular dynamics simulations are able to satisfactorily classify all seven mutations in our clinically derived testset. A result for a single mutation could therefore be returned in less than an hour, thereby demonstrating that this or similar methods are now sufficiently fast and reproducible for clinical use.

A Molecular Dynamics-Shared Pharmacophore Approach to Boost Early-Enrichment Virtual Screening: A Case Study on Peroxisome Proliferator-Activated Receptor α.

Nuclear magnetic resonance (NMR) provides structural and dynamic information reflecting an average, often non-linear, of multiple solution-state conformations. Therefore, a single optimized structure derived from NMR refinement may be misleading if the NMR data actually result from averaging of distinct conformers. It is hypothesized that a conformational ensemble generated by a valid molecular dynamics (MD) simulation should be able to improve agreement with the NMR data set compared with the single optimized starting structure. Using a model system consisting of two sequence-related self-complementary ribonucleotide octamers for which NMR data was available, 0.3 ns particle mesh Ewald MD simulations were performed in the AMBER force field in the presence of explicit water and counterions. Agreement of the averaged properties of the molecular dynamics ensembles with NMR data such as homonuclear proton nuclear Overhauser effect (NOE)-based distance constraints, homonuclear proton and heteronuclear (1)H-(31)P coupling constant (J) data, and qualitative NMR information on hydrogen bond occupancy, was systematically assessed. Despite the short length of the simulation, the ensemble generated from it agreed with the NMR experimental constraints more completely than the single optimized NMR structure. This suggests that short unrestrained MD simulations may be of utility in interpreting NMR results. As expected, a 0.5 ns simulation utilizing a distance dependent dielectric did not improve agreement with the NMR data, consistent with its inferior exploration of conformational space as assessed by 2-D RMSD plots. Thus, ability to rapidly improve agreement with NMR constraints may be a sensitive diagnostic of the MD methods themselves.

Viral adhesion to host cells is a critical step in infection for many viruses, including monkeypox virus (MPXV). In MPXV, the H3 protein mediates viral adhesion through its interaction with heparan sulfate (HS), yet the structural details of this interaction have remained elusive. Using AI-based structural prediction tools and molecular dynamics (MD) simulations, we identified a novel, positively charged α-helical domain in H3 that is essential for HS binding. This conserved domain, found across orthopoxviruses, was experimentally validated and shown to be critical for viral adhesion, making it an ideal target for antiviral drug development. Targeting this domain, we designed a protein inhibitor, which disrupted the H3-HS interaction, inhibited viral infection in vitro and viral replication in vivo, offering a promising antiviral candidate. Our findings reveal a novel therapeutic target of MPXV, demonstrating the potential of combination of AI-driven methods and MD simulations to accelerate antiviral drug discovery.