Molecular Dynamics Trajectories Research Articles

A simple approach to rotationally invariant machine learning of a vector quantity.

Unlike with the energy, which is a scalar property, machine learning (ML) prediction of vector or tensor properties poses the additional challenge of achieving proper invariance (covariance) with respect to molecular rotation. For the energy gradients needed in molecular dynamics (MD), this symmetry is automatically fulfilled when taking analytic derivative of the energy, which is a scalar invariant (using properly invariant molecular descriptors). However, if the properties cannot be obtained by differentiation, other appropriate methods should be applied to retain the covariance. Several approaches have been suggested to properly treat this issue. For nonadiabatic couplings and polarizabilities, for example, it was possible to construct virtual quantities from which the above tensorial properties are obtained by differentiation and thus guarantee the covariance. Another possible solution is to build the rotational equivariance into the design of a neural network employed in the model. Here, we propose a simpler alternative technique, which does not require construction of auxiliary properties or application of special equivariant ML techniques. We suggest a three-step approach, using the molecular tensor of inertia. In the first step, the molecule is rotated using the eigenvectors of this tensor to its principal axes. In the second step, the ML procedure predicts the vector property relative to this orientation, based on a training set where all vector properties were in this same coordinate system. As the third step, it remains to transform the ML estimate of the vector property back to the original orientation. This rotate-predict-rotate (RPR) procedure should thus guarantee proper covariance of a vector property and is trivially extensible also to tensors such as polarizability. The RPR procedure has an advantage that the accurate models can be trained very fast for thousands of molecular configurations, which might be beneficial where many training sets are required (e.g., in active learning). We have implemented the RPR technique, using the MLatom and Newton-X programs for ML and MD, and performed its assessment on the dipole moment along MD trajectories of 1,2-dichloroethane.

Structural and functional consequences of non-synonymous SNPs within the LAMA2 protein: a molecular dynamics perspective

Clinical phenotypic presentations associated with LAMA2 deficiency have shown a variety of manifestations. LAMA2 mutations are mainly linked to congenital muscular dystrophy, but there is also mounting evidence suggesting their presence in inflammatory breast cancer, laryngopharyngeal squamous cell carcinoma, and ventricular tachycardia related to coronary artery disease and cardiomyopathy. This study examined the structural and functional impacts of 144 non-synonymous single nucleotide polymorphisms (nsSNPs) within the LAMA2 gene. Through multi-tiered sequence and structure-based methods, 11 deleterious and destabilizing mutations were identified (A1362T, E1308Q, E1360G, I1276S, L1195P, M1359T, P1232H, P1238A, P1272L, Y1234H, Y1338C). Further, four mutations (L1195P, Y1234H, P1238A, A1362T), which aligned with conserved positions, were subjected to 500 ns molecular dynamics (MD) simulations. RMSD calculated from MD trajectories highlighted structural disparities between wild-type and mutant forms, with the latter showing greater flexibility. Radius of gyration analysis indicated reduced compactness, solvent accessibility changes suggested unfolding, and hydrogen bond (HB) analysis demonstrated disrupted integrity. The HB analysis revealed disruptions in structural integrity due to diminished hydrogen bonds in mutants. Secondary structure analysis revealed significant alterations in secondary structural content. Principal Component Analysis unveiled increased dynamic behavior in mutants. Gibbs free energy landscape analysis reflected distinct energy minima regions in mutants, indicating structural destabilization. Overall, this study revealed the functional and structural ramifications of nsSNPs in the LAMA2 gene, providing valuable insights into potential disease-causing mutations and warranting future research on understanding LAMA2 associated diseases and disorders.

Explicit Configurational Entropy of Mixing in Molecular Dynamics Simulations.

The entropy of mixing of a multicomponent system of particles is a simple expression of the molar fractions for the equilibrium state, but its intermediate values for transient (nonequilibrium) states can not be calculated directly from the particle coordinates so far. We propose a simple expression for the configurational entropy of mixing based solely on the set of instantaneous coordinates, which is suitable for the on-the-fly determination of the degree of mixing along a molecular dynamics trajectory. We illustrate the applicability of our scheme with the example of several molecular mixtures that exhibit fast and slow mixing and demixing processes within a molecular dynamics simulation.

Inferring the existence of hydrogen bonds directly from statistical analysis of molecular dynamics trajectories.

As a demonstration of how fundamental chemical concepts can be gleaned from data using machine learning methods, we demonstrate the automated detection of hydrogen bonds by statistical analysis of molecular dynamics trajectories. In particular, we infer the existence and nature of electrostatically driven noncovalent interactions by examining the relative probability of supramolecular configurations with and without electrostatic interactions. Then, using Laplacian eigenmaps clustering, we identify hydrogen bonding motifs in hydrogen fluoride, water, and methanol. The hydrogen bonding motifs that we identify support traditional geometric criteria.

StreaMD: the toolkit for high-throughput molecular dynamics simulations.

Molecular dynamics simulations serve as a prevalent approach for investigating the dynamic behaviour of proteins and protein-ligand complexes. Due to its versatility and speed, GROMACS stands out as a commonly utilized software platform for executing molecular dynamics simulations. However, its effective utilization requires substantial expertise in configuring, executing, and interpreting molecular dynamics trajectories. Existing automation tools are constrained in their capability to conduct simulations for large sets of compounds with minimal user intervention, or in their ability to distribute simulations across multiple servers. To address these challenges, we developed a Python-based tool that streamlines all phases of molecular dynamics simulations, encompassing preparation, execution, and analysis. This tool minimizes the required knowledge for users engaging in molecular dynamics simulations and can efficiently operate across multiple servers within a network or a cluster. Notably, the tool not only automates trajectory simulation but also facilitates the computation of free binding energies for protein-ligand complexes and generates interaction fingerprints across the trajectory. Our study demonstrated the applicability of this tool on several benchmark datasets. Additionally, we provided recommendations for end-users to effectively utilize the tool.Scientific contributionThe developed tool, StreaMD, is applicable to different systems (proteins, ligands and their complexes including co-factors) and requires a little user knowledge to setup and run molecular dynamics simulations. Other features of StreaMD are seamless integration with calculation of MM-GBSA/PBSA binding free energies and protein-ligand interaction fingerprints, and running of simulations within distributed environments. All these will facilitate routine and massive molecular dynamics simulations.

Virtual screening of plant-derived molecules against zinc-dependent imipenemases in class B metallo-β-lactamases of Acinetobacter baumannii.

Metallo-β-lactamases (MBLs), enzymes of class B, employ zinc ions to degrade β-lactam antibiotics such as penicillins, cephalosporins, carbapenems, and cephamycins. Carbapenem-resistant Acinetobacter baumannii (CRAB) is linked to the existence of carbapenemase enzymes such as oxacillinase and MBL. The most prevalent resistance mechanisms include imipenemases (IMP), verona integron-encoded MBL, and New Delhi MBL-1. The effectiveness of current antibiotics against the MBL enzyme is limited due to the presence of metal ions, underscoring the need for new antimicrobial agents. Recent research has demonstrated that natural compounds can effectively inhibit MBL. This study aims to screen natural phytochemicals against IMP-2 MBL using in silico virtual screening techniques via AutoDock Vina and molecular dynamic simulations with GROMACS for 200ns, followed by molecular mechanics/Poisson‒Boltzmann surface area analysis. This procedure identified new lead molecules against A. baumannii that produce IMP. A total of 588 natural compounds were screened against IMP, along with the imipenem substrate and known inhibitors of L-captopril. The top four compounds, N025-0038 (NC1), N062-0008 (NC2), eupalitin, and Rosmorinic acid, demonstrated binding affinities of ‒8.5, ‒8.4, ‒7.5, and ‒7.2kcal/mol, respectively. The structural stability of these complexes was observed to be maintained throughout the simulation in a dynamic environment, as determined by molecular dynamics trajectory analysis, and all these compounds met the SWISS-ADME (adsorption, distribution, metabolism, and excretion) properties. NC1 and NC2 compounds are considered potential drug molecules against IMP. However, while these selected compounds showed superior binding energy in computational analysis, further in vitro analysis is required to establish an effective drug regimen against A. baumannii that produces IMP.

Critical assessment of diffusion coefficients of benzene and its derivatives in supercritical carbon dioxide: Experimental data, molecular dynamics simulations, and modeling

The tracer diffusion coefficient (D12) is a critical transport property essential for research, design and optimization of industrial processes. However, experimental determination of D12 is often challenging, time-consuming, and frequently exhibits significant variability across different systems and conditions. In this study, a comprehensive assessment of D12 for benzene, phenol, chlorobenzene, anisole, and the three isomers of dichlorobenzene in supercritical carbon dioxide (▪) is presented. The available experimental data was meticulously compiled and analyzed, comparing different data points from various authors to assess consistency and accuracy. The experimental values were then compared with predictions from classical molecular dynamics (MD) simulations carried out in this work, in which three distinct ▪ force fields (EPM2, TraPPE, and Zhu et al.) and the OPLS-AA force field for the solute molecules were tested in combination with NVT and NPT ensembles. Among the evaluated combinations of force field and ensemble, the Zhu et al./NPT pair demonstrated excellent agreement with experimental data, achieving an overall deviation of only 5.63 %.Additionally, the performance of alternative approaches for estimating D12, including phenomenological and semi-empirical models—Wilke-Chang, Lai-Tan, Tracer Liu-Silva-Macedo and He-Yu equations—and machine learning (ML) algorithms was evaluated, highlighting their comparative accuracy while accounting for their inherent limitations. Overall, the ML model proved highly effective, with a deviation of only 2.42 % when applied to compounds within the training dataset. However, for compounds outside this dataset, the reliability, flexibility and accuracy of MD simulations, as demonstrated in this study, strongly suggest their application for D12 estimation in ▪/benzene-derivative systems, with possible additional valuable insights from the analysis of MD trajectories.

Determination of the Dipole Moment Variation Upon Excitation in the Chromophore of Green Fluorescent Protein From Molecular Dynamic Trajectories with QM/MM Potentials Using Machine Learning Methods

Quantum and molecular mechanics (QM/MM) potentials are used to calculate molecular dynamics trajectories for the EYFP protein of the green fluorescent protein family. Machine learning models are constructed to establish the relationship between the geometric parameters of the chromophore in the frame of its trajectory and the properties of its electronic excitation. It is shown that it is not enough to use only bridging bonds between the phenyl and imidazolidone fragments of the chromophore as a geometric parameter, and at least two more neighboring bonds must be added to the model. The proposed models allow determination of the dipole moment variation upon excitation with an average error of 0.11 a.u.

PIM-1L Kinase Binds to and Inactivates SRPK1: A Biochemical and Molecular Dynamics Study.

SR/RS dipeptide repeats vary in both length and position, and are phosphorylated by SR protein kinases (SRPKs). PIM-1L, the long isoform of PIM-1 kinase, the splicing of which has been implicated in acute myeloid leukemia, contains a domain that consists largely of repeating SR/RS and SH/HS dipeptides (SR/SH-rich). In order to extend our knowledge on the specificity and cellular functions of SRPK1, here we investigate whether PIM-1L could act as substrate of SRPK1 by a combination of biochemical and computational approaches. Our biochemical data showed that the SR/SH-rich domain of PIM-1L was able to associate with SRPK1, yet it could not act as a substrate but, instead, inactivated the kinase. In line with our biochemical data, molecular modeling followed by a microsecond-scale all-atom molecular dynamics (MD) simulation suggests that the SR/SH-rich domain acts as a pseudo-docking peptide that binds to the same acidic docking-groove used in other SRPK1 interactions and induces inactive SRPK1 conformations. Comparative community network analysis of the MD trajectories, unraveled the dynamic architecture of apo SRPK1 and notable alterations of allosteric communications upon PIM-1L peptide binding. This analysis also allowed us to identify key SRPK1 residues, including unique ones, with a pivotal role in mediating allosteric signal propagation within the kinase core. Interestingly, most of the identified amino acids correspond to cancer-associated amino acid changes, validating our results. In total, this work provides insights not only on the details of SRPK1 inhibition by the PIM-1L SR/SH-domain, but also contributes to an in-depth understanding of SRPK1 regulation.

Molecular Docking Assessment of Limonoids from Cameroonian Entandrophragma Species as Potential Inhibitors of Anopheles gambiae Acetylcholinesterase (AChE)

Malaria remains one of the great killers in tropical regions of the world due to the transmission of the Plasmodium parasite by the bites of the female mosquito Anopheles. The resistance of this species to synthetic insecticides contributes to an increase in the incidence of malaria and therefore necessitates the development of new potent and eco-friendly insecticides. In this study, twelve previously reported limonoids from four Entandrophragma species collected in Cameroon have been computationally evaluated for their Anopheles gambiae AChE inhibitory activity. The docking procedure was carried out through Molecular Operating Environment 2019.01 (MOE), while the UCSF Chimera program was used to model the docking results based on interactions between proteins and ligands, and molecular dynamics trajectories were analyzed using the GROMACS 2021.1 tool. Entandrophragmin and encandollens B and C with docking scores ranging from −6.45 to −7.28 kcal/mol were the most promising hits compared to the reference azadirachtin (−6.22 kcal/mol) and were further evaluated for their mechanism of action. Subsequent evaluation classified encandollen C as the best candidate for the development of new potent eco-friendly insecticides based on its lower average RMSD and RMSF and its compactness over a 150 ns duration with acetylcholinesterase.

Phenol Photostatic Spectra and Quantum‐Classical Photodynamic Deprotonation

ABSTRACTThe spectral‐luminescence properties and photochemical conversions of phenol were analyzed for an isolated molecule as well as in water solvents in a continuum implicit model and explicit atomistic surroundings. This involved employing cut‐edge hybrid quantum‐classical methodologies to generate static optical spectra and the excited dissipative crossing potential energy curves. A combination of electronic excitations, gradient calculations, and embedding electrostatic potential fitting charges on quantum‐classical molecular dynamic propagation trajectories provided statistically averaged absorption spectra. The mixed‐reference spin‐flip multiconfigurational linear response method based on reference triplet preprocessed in the time‐dependent density‐functional theory was utilized to determine conical intersections between the lowest excited and ground states, as well as two‐stage transitions from the second excitation to the ground state. Non‐adiabatic quantum‐classical molecular dynamics defined photodissipative trajectories of excited states, their lifetimes, and crossing points through trajectory surface hopping together with the mixed‐reference spin‐flip and embedding electrostatic potential fitting approaches. Dyson orbitals of the extended Koopmans' theorem were applied to reveal the nature of molecular states at conical intersections and key points on photodynamic trajectories. Potential hydroxyl group cleavage predicted with conical intersections searching turns to “swift” OH deprotonation through |π→⟩ transition along photodynamic propagations in contrast with “long” processes leading to benzene ring deformation with stable OH bond.

Molecular Dynamics (MD) Simulations Provide Insights into the Activation Mechanisms of 5-HT2A Receptors

Recent breakthroughs in the determination of atomic resolution 3-D cryo-electron microscopy structures of membrane proteins present an unprecedented opportunity for drug discovery. Structure-based drug discovery utilizing in silico methods enables the study of dynamic connectivity of stable conformations induced by the drug in achieving its effect. With the ever-expanding computational power, simulations of this type reveal protein dynamics in the nano-, micro-, and even millisecond time scales. In the present study, aiming to characterize the protein dynamics of the 5HT2A receptor stimulated by ligands (agonist/antagonist), we performed 1 µs MD simulations on 5HT2A/DOI (agonist), 5HT2A/GSK215083 (antagonist), and 5HT2A (APO, no ligand) systems. The crystal structure of 5HT2A/zotepine (antagonist) (PDB: 6A94) was used to set up the simulation systems in a lipid bilayer environment. We found the monitoring of the ionic lock residue pair (R3.50-E6.30) of 5HT2A in MD simulations to be a good approximation of the effects of agonists (ionic lock breakage) or antagonists (ionic lock formation) on receptor activation. We further performed analyses of the MD trajectories, including Principal Component Analysis (PCA), hydrogen bond, salt bridge, and hydrophobic interaction network analyses, and correlation between residues to identify key elements of receptor activation. Our results suggest that in order to trigger receptor activation, DOI must interact with 5HT2A through residues V5.39, G5.42, S5.43, and S5.46 on TM5, inducing significant conformational changes in the backbone angles of G5.42 and S5.43. DOI also interacted with residues W6.48 (toggle switch) and F6.51 on TM6, causing major conformational shifts in the backbone angles of F6.44 and V6.45. These structural changes were transmitted to the intracellular ends of TM5, TM6, and ICL3, resulting in the breaking of the ionic lock and subsequent G protein activation. The studies could be helpful in future design of selective agonists/antagonists for various serotonin receptors (5HT1A, 5HT2A, 5HT2B, 5HT2C, and 5HT7) involved in detrimental disorders, such as addiction and schizophrenia.

Two riboswitch classes that share a common ligand-binding fold show major differences in the ability to accommodate mutations.

Riboswitches are structured RNAs that sense small molecules to control expression. Prequeuosine1 (preQ1)-sensing riboswitches comprise three classes (I, II and III) that adopt distinct folds. Despite this difference, class II and III riboswitches each use 10 identical nucleotides to bind the preQ1 metabolite. Previous class II studies showed high sensitivity to binding-pocket mutations, which reduced preQ1 affinity and impaired function. Here, we introduced four equivalent mutations into a class III riboswitch, which maintained remarkably tight preQ1 binding.Co-crystal structures of each class III mutant showed compensatory interactions that preserve the fold. Chemical modification analysis revealed localized RNA flexibility changes for each mutant, but molecular dynamics (MD) simulations suggested that each mutation was not overtly destabilizing. Although impaired, class III mutants retained tangible gene-regulatory activity in bacteria compared to equivalent preQ1-II variants; mutations in the preQ1-pocket floor were tolerated better than wall mutations. Principal component analysis of MD trajectories suggested that the most functionally deleterious wall mutation samples different motions compared to wildtype. Overall, the results reveal that formation of compensatory interactions depends on the context of mutations within the overall fold and that functionally deleterious mutations can alter long-range correlated motions that link the riboswitch binding pocket with distal gene-regulatory sequences.

SeqDance: A Protein Language Model for Representing Protein Dynamic Properties.

Proteins perform their functions by folding amino acid sequences into dynamic structural ensembles. Despite the important role of protein dynamics, their complexity and the absence of efficient representation methods have limited their integration into studies on protein function and mutation fitness, especially in deep learning applications. To address this, we present SeqDance, a protein language model designed to learn representation of protein dynamic properties directly from sequence alone. SeqDance is pre-trained on dynamic biophysical properties derived from over 30,400 molecular dynamics trajectories and 28,600 normal mode analyses. Our results show that SeqDance effectively captures local dynamic interactions, co-movement patterns, and global conformational features, even for proteins lacking homologs in the pre-training set. Additionally, we showed that SeqDance enhances the prediction of protein fitness landscapes, disorder-to-order transition binding regions, and phase-separating proteins. By learning dynamic properties from sequence, SeqDance complements conventional evolution- and static structure-based methods, offering new insights into protein behavior and function.

Rare-Event Sampling using a Reinforcement Learning-Based Weighted Ensemble Method.

Despite the power of path sampling strategies in enabling simulations of rare events, such strategies have not reached their full potential. A common challenge that remains is the identification of a progress coordinate that captures the slow relevant motions of a rare event. Here we have developed a weighted ensemble (WE) path sampling strategy that exploits reinforcement learning to automatically identify an effective progress coordinate among a set of potential coordinates during a simulation. We apply our WE strategy with reinforcement learning to three benchmark systems: (i) an egg carton-shaped toy potential, (ii) an S-shaped toy potential, and (iii) a dimer of the HIV-1 capsid protein (C-terminal domain). To enable rapid testing of the latter system at the atomic level, we employed discrete-state synthetic molecular dynamics trajectories using a generative, fine-grained Markov state model that was based on extensive conventional simulations. Our results demonstrate that using concepts from reinforcement learning with a weighted ensemble of trajectories automatically identifies relevant progress coordinates among multiple candidates at a given time during a simulation. Due to the rigorous weighting of trajectories, the simulations maintain rigorous kinetics.

Application of the flexible and polarisable hydration ion model to study Th(IV) aqueous solutions

Th(IV) hydration in aqueous solution is studied by means of molecular dynamics simulations based on a polarisable Th(IV)- H 2 O interaction potential developed in the framework of the hydrated ion concept. Molecular Dynamics (MD) simulations provide as the only in-solution representative motif the ennea-aqua ion, [ Th ( H 2 O ) 9 ] 4 + , with first shell water molecules at an average distance of 2.47 Å from the ion, in accordance with most of the experimental estimates. A well-defined second hydration shell is also identified, hosting ca. 19 solvent molecules at 4.65 Å. Non-structural aspects of the Th(IV) hydration phenomenon, such hydration enthalpy and ion diffusion are well reproduced by the simulation results. The analysis of the O-Th-O angle distribution suggests that the capped square antiprism arrangement prevails over the trigonal tricapped prism. The well-balanced definition of the ion–water and water–water interactions in such a demanding ion neighbourhood is confirmed by the EXAFS and XANES spectroscopies. The theoretical spectra, computed from the MD trajectory, are in fine agreement with some of the experimentally recorded. The sensitivity of the XAS spectroscopy to structural changes confirms how the presented potential manages in a proper way the response of the water molecules to the highly polarising electric field generated by the tetravalent ion.

Computational insights into dispiro-oxindolopyrrolizidines as aldose reductase inhibitors: Molecular docking, ADME analysis, molecular dynamic simulation, and MM-PBSA analysis

Chronic diabetes problems arise from the activation of aldose reductase (AR), the primary enzyme of the polyol pathway, under hyperglycemic circumstances. The spirooxindole ring system comes out to be a fascinating class of naturally occurring compounds with pharmaceutical interest. In this work, we have chosen three dispirooxindolopyrrolizidines (DSOIP-H, DSOIP-Me, and DSOIP-OMe) to investigate their ability to inhibit AR. Through molecular docking studies, the order of binding energy was calculated as DSOIP-H (-8.23 kcal/mol) < DSOIP-Me (-8.58 kcal/mol) < DSOIP-OMe (-8.72 kcal/mol). The three ligands showed effective binding with AR's catalytic active site through hydrophobic interactions. The ADME analysis revealed valuable insights into the ligands' pharmacokinetic properties. Through protein-ligand interactions, DSOIP-OMe consistently interacted with the amino acid residues of the AR enzyme's active site areas, as demonstrated by the molecular dynamic (MD) trajectory analysis. The exceptional inhibitory efficacy of DSOIP-OMe has been established based on its binding energy values of -92.69 kJ/mol as determined by MM-PBSA. The PHE122 residue of AR is recognized as a crucial residue that made a substantial contribution to the binding energy of MM-PBSA through hydrophobic interactions.

Dy-SheHeRASADe: A representation of the [formula omitted] sheet dynamics through surface descriptors

Molecular dynamics (MD) simulations are important tools for studying the dynamic motions of macromolecules at the atomic level. With the increasing capabilities of high performance computing, MD simulations are becoming more widely used. This allows molecular modelers to simulate the molecular behavior of large molecular architectures for much longer trajectories. Appropriate visualization of MD trajectories is becoming essential to provide an immediate and intuitive understanding of a molecule’s dynamics and function. In this study, we implement a novel 3D graphical representation, Dynamical Sheets Helper for RepresentAtion of SurfAce Descriptors (Dy-SheHeRASADe), to visualize the β sheet secondary structures of proteins in the context of molecular dynamics. Dy-SheHeRASADe is developed in UnityMol, an open source molecular viewer and prototyping platform. We considered β sheet fluctuations and hydrogen bond formation during molecular dynamics simulations to characterize the parts of β sheets with large motions or with labile bonds. We propose two visualization modes based on a surface representation of the β sheets calculated according to the positions of the α carbons and the hydrogen bonds between the β strands. The volumetric mode, in which this surface is enclosed in a semi-transparent volume that represents the fluctuation zone of the sheet during dynamics. The heatmap mode, in which the surface is colored according to the amplitude values of the α carbons. In addition, we quantify the β sheet fluctuations by displaying the values of the largest and smallest movements of the β sheets, the surface area of the sheets, and the number of hydrogen bonds.

Overcoming methicillin resistance by methicillin-resistant Staphylococcus aureus: Computational evaluation of napthyridine and oxadiazoles compounds for potential dual inhibition of PBP-2a and FemA proteins

Abstract The treatment of the various infections caused by Staphylococcus aureus has become challenging due to the evolving resistance against current therapeutics. In this study, the potentials of napthyridine and oxadiazole derivatives to serve as dual inhibitors of penicillin-binding protein 2a (PBP-2a) and FemA protein, which are crucial to resistance to methicillin-based drugs by S. aureus, were evaluated using molecular modeling techniques. Seventy-two compounds were subjected to molecular docking against the proteins, and the hit compounds were subjected to drug-likeness evaluation and in silico pharmacokinetics prediction. The compounds with good safety profiles were subjected to a 250-ns molecular dynamics (MD) simulation and other relevant analyses based on the MD trajectories. Five hit compounds were selected based on their high affinity for the targets as evidenced by their docking scores ranging from −8.6 to −10.1 kcal/mol for PBP-2a and −9.6 to −9.9 kcal/mol for FemA. These compounds also passed Lipinski’s rule of five evaluation with no violation and possessed high human intestinal absorption potential, showcasing their potential as orally administered therapeutic agents. However, three of the compounds were potential mutagens. MD simulation revealed that the final two compounds maintained stable interactions with the target proteins over 250 ns, with minimal deviations and fluctuations. Hydrogen bond stability and energy decomposition analysis further confirmed the strong binding affinity of the hit compounds compared to the control drug, methicillin. Conclusively, the compounds with the CID “135964525” and “44130718” are worthy of further experimental validation in the development of potential inhibitors of PBP-2a and FemA.

Graph Neural Network-Based Molecular Property Prediction with Patch Aggregation.

Graph neural networks (GNNs) have emerged as powerful tools for quantum chemical property prediction, leveraging the inherent graph structure of molecular systems. GNNs depend on an edge-to-node aggregation mechanism for combining edge representations into node representations. Unfortunately, existing learnable edge-to-node aggregation methods substantially increase the number of parameters and, thus, the computational cost relative to simple sum aggregation. Worse, as we report here, they often fail to improve predictive accuracy. We therefore propose a novel learnable edge-to-node aggregation mechanism that aims to improve the accuracy and parameter efficiency of GNNs in predicting molecular properties. The new mechanism, called "patch aggregation", is inspired by the Multi-Head Attention and Mixture of Experts machine learning techniques. We have incorporated the patch aggregation method into the specialized, state-of-the-art GNN models SchNet, DimeNet++, SphereNet, TensorNet, and VisNet and show that patch aggregation consistently outperforms existing learnable and nonlearnable aggregation techniques (sum, multilayer perceptron, softmax, and set transformer aggregation) in the prediction of molecular properties such as QM9 thermodynamic properties and MD17 molecular dynamics trajectory energies and forces. We also find that patch aggregation not only improves prediction accuracy but also is parameter-efficient, making it an attractive option for practical applications for which computational resources are limited. Further, we show that Patch aggregation can be applied across different GNN models. Overall, Patch aggregation is a powerful edge-to-node aggregation mechanism that improves the accuracy of molecular property predictions by GNNs.