MHC-Fine: Fine-tuned AlphaFold for Precise MHC-Peptide Complex Prediction.

The precise prediction of Major Histocompatibility Complex (MHC)-peptide complex structures is pivotal for understanding cellular immune responses and advancing vaccine design. In this study, we enhanced AlphaFold's capabilities by fine-tuning it with a specialized dataset consisting of exclusively high-resolution Class I MHC-peptide crystal structures. This tailored approach aimed to address the generalist nature of AlphaFold's original training, which, while broad-ranging, lacked the granularity necessary for the high-precision demands of Class I MHC-peptide interaction prediction. A comparative analysis was conducted against the homology-modeling-based method Pandora, as well as the AlphaFold multimer model. Our results demonstrate that our fine-tuned model outperforms others in terms of RMSD (median value for C-alpha atoms for peptides is 0.66 Å) and also provides enhanced predicted lDDT scores, offering a more reliable assessment of the predicted structures. These advances have substantial implications for computational immunology, potentially accelerating the development of novel therapeutics and vaccines by providing a more precise computational lens through which to view MHC-peptide interactions.

k-Means NANI: an improved clustering algorithm for Molecular Dynamics simulations.

One of the key challenges of k-means clustering is the seed selection or the initial centroid estimation since the clustering result depends heavily on this choice. Alternatives such as k-means++ have mitigated this limitation by estimating the centroids using an empirical probability distribution. However, with high-dimensional and complex datasets such as those obtained from molecular simulation, k-means++ fails to partition the data in an optimal manner. Furthermore, stochastic elements in all flavors of k-means++ will lead to a lack of reproducibility. K-means N-Ary Natural Initiation (NANI) is presented as an alternative to tackle this challenge by using efficient n-ary comparisons to both identify high-density regions in the data and select a diverse set of initial conformations. Centroids generated from NANI are not only representative of the data and different from one another, helping k-means to partition the data accurately, but also deterministic, providing consistent cluster populations across replicates. From peptide and protein folding molecular simulations, NANI was able to create compact and well-separated clusters as well as accurately find the metastable states that agree with the literature. NANI can cluster diverse datasets and be used as a standalone tool or as part of our MDANCE clustering package.

Exact analytical algorithm for solvent accessible surface area and derivatives in implicit solvent molecular simulations on GPUs.

In this paper, we present dSASA (differentiable SASA), an exact geometric method to calculate solvent accessible surface area (SASA) analytically along with atomic derivatives on GPUs. The atoms in a molecule are first assigned to tetrahedra in groups of four atoms by Delaunay tetrahedrization adapted for efficient GPU implementation and the SASA values for atoms and molecules are calculated based on the tetrahedrization information and inclusion-exclusion method. The SASA values from the numerical icosahedral-based method can be reproduced with more than 98% accuracy for both proteins and RNAs. Having been implemented on GPUs and incorporated into the software Amber, we can apply dSASA to implicit solvent molecular dynamics simulations with inclusion of this nonpolar term. The current GPU version of GB/SA simulations has been accelerated up to nearly 20-fold compared to the CPU version, outperforming LCPO, a commonly used, fast algorithm for calculating SASA, as the system size increases. While we focus on the accuracy of the SASA calculations for proteins and nucleic acids, we also demonstrate stable GB/SA MD mini-protein simulations.

MHC-Fine: Fine-tuned AlphaFold for Precise MHC-Peptide Complex Prediction.

The precise prediction of Major Histocompatibility Complex (MHC)-peptide complex structures is pivotal for understanding cellular immune responses and advancing vaccine design. In this study, we enhanced AlphaFold's capabilities by fine-tuning it with a specialized dataset comprised by exclusively high-resolution MHC-peptide crystal structures. This tailored approach aimed to address the generalist nature of AlphaFold's original training, which, while broad-ranging, lacked the granularity necessary for the high-precision demands of MHC-peptide interaction prediction. A comparative analysis was conducted against the homology-modeling-based method Pandora [13], as well as the AlphaFold multimer model [8]. Our results demonstrate that our fine-tuned model outperforms both in terms of RMSD (median value is 0.65 Å) but also provides enhanced predicted lDDT scores, offering a more reliable assessment of the predicted structures. These advances have substantial implications for computational immunology, potentially accelerating the development of novel therapeutics and vaccines by providing a more precise computational lens through which to view MHC-peptide interactions.

AmberTools.

AmberTools is a free and open-source collection of programs used to set up, run, and analyze molecular simulations. The newer features contained within AmberTools23 are briefly described in this Application note.

Rapid Rescoring and Refinement of Ligand-Receptor Complexes Using Replica Exchange Molecular Dynamics with a Monte Carlo Pose Reservoir.

Virtual screening (VS) involves generation of poses for a library of ligands and ranking using simplified energy functions and limited flexibility. Top-scored poses are used to rank and prioritize ligands. Here, we adapt the reservoir replica exchange molecular dynamics (res-REMD) method to rerank poses generated through VS. REMD simulations are carried out but with occasional Monte Carlo jumps to alternate VS-generated poses using a Metropolis criterion. The simulations converge within 10 ns for all systems, generating populations of alternate poses in the context of fully flexible ligand and protein side chains. The protocol is applied to four model protein-ligand complexes, where DOCK resulted in two successes and two scoring failures. In all four systems, the most populated cluster from the final ensemble exhibits high similarity to the crystallographic pose with ligand RMSD values under 2.0 Å. Both DOCK failures were rescued. For one DOCK success, the protocol identified the correct pose but also sampled an alternate pose at equal probability. Opportunities for future improvements and extensions are discussed.

Exploring the Transferability of Replica Exchange Structure Reservoirs to Accelerate Generation of Ensembles for Alternate Hamiltonians or Protein Mutations.

Generating precise ensembles is commonly a prerequisite to understand the energetics of biological processes using Molecular Dynamics (MD) simulations. Previously, we have shown how unweighted reservoirs built from high temperature MD simulations can accelerate convergence of Boltzmann-weighted ensembles by at least 10× with the Reservoir Replica Exchange MD (RREMD) method. Therefore, in this work, we explore whether an unweighted structure reservoir generated with one Hamiltonian (solute force field plus solvent model) can be reused to quickly generate accurately weighted ensembles for Hamiltonians other than the one that was used to generate the reservoir. We also extended this methodology to rapidly estimate the effects of mutations on peptide stability by using a reservoir of diverse structures obtained from wild-type simulations. These results suggest that structures generated via fast methods such as coarse-grained models or structures predicted by Rosetta or deep learning approaches could be integrated into a reservoir to accelerate generation of ensembles using more accurate representations.

SpikeScape: A Tool for Analyzing Structural Diversity in Experimental Structures of the SARS-CoV-2 Spike Glycoprotein.

In this application note, we describe a tool which we developed to help structural biologists who study the SARS-CoV-2 spike glycoprotein. There are more than 500 structures of this protein available in the Protein Data Bank. These structures are available in different flavors: wild type spike, different variants, 2P substitutions, structures with bound antibodies, structures with Receptor Binding Domains (RBD) in closed or open conformation, etc. Understanding differences between these structures could provide insight into how the spike structure changes in different variants or upon interaction with different molecules such as receptors or antibodies. However, inconsistencies among deposited structures, such as different chain or sequence numbering, hamper a straightforward comparison of all structures. The tool described in this note fixes those chain inconsistencies and calculates the distribution of the requested distance between any two atoms across all SARS-CoV-2 spike structures available in the Protein Data Bank (excluding PDB files with only spike fragments such as the RBD), with the option to filter by various selections. The tool provides a histogram and cumulative frequency of the calculated distribution, as the ability to download the results and corresponding PDB IDs.

On the plasticity of amyloid formation: The impact of destabilizing small to large substitutions on islet amyloid polypeptide amyloid formation.

Amyloids are partially ordered, proteinaceous, ß-sheet rich deposits that have been implicated in a wide range of diseases. An even larger set of proteins that do not normally form amyloid in vivo can be induced to do so in vitro. A growing number of structures of amyloid fibrils have been reported and a common feature is the presence of a tightly packed core region in which adjacent monomers pack together in extremely tight interfaces, often referred to as steric zippers. A second common feature of many amyloid fibrils is their polymorphous nature. We examine the consequences of disrupting the tight packing in amyloid fibrils on the kinetics of their formation using the 37 residue polypeptide hormone islet amyloid polypeptide (IAPP, amylin) as a model system. IAPP forms islet amyloid in vivo and is aggressively amyloidogenic in vitro. Six Cryo-EM structures of IAPP amyloid fibrils are available and in all Gly24 is in the core of the structured region and makes tight contacts with other residues. Calculations using the ff14SBonlysc forcefield in Amber20 show that substitutions with larger amino acids significantly disrupt close packing and are predicted to destabilize the various fibril structures. However, Gly to 2-amino butyric acid (2-carbon side chain) and Gly to Leu substitutions actually enhance the rate of amyloid formation. A Pro substitution slows, but does not prevent amyloid formation.

BioMThermDB 1.0: Thermophysical Database of Proteins in Solutions.

We present here a freely available web-based database, called BioMThermDB 1.0, of thermophysical and dynamic properties of various proteins and their aqueous solutions. It contains the hydrodynamic radius, electrophoretic mobility, zeta potential, self-diffusion coefficient, solution viscosity, and cloud-point temperature, as well as the conditions for those determinations and details of the experimental method. It can facilitate the meta-analysis and visualization of data, can enable comparisons, and may be useful for comparing theoretical model predictions with experiments.

SARS-CoV-2 spike conformation determines plasma neutralizing activity elicited by a wide panel of human vaccines.

Numerous safe and effective coronavirus disease 2019 vaccines have been developed worldwide that use various delivery technologies and engineering strategies. We show here that vaccines containing prefusion-stabilizing S mutations elicit antibody responses in humans with enhanced recognition of S and the S1 subunit relative to postfusion S as compared with vaccines lacking these mutations or natural infection. Prefusion S and S1 antibody binding titers positively and equivalently correlated with neutralizing activity, and depletion of S1-directed antibodies completely abrogated plasma neutralizing activity. We show that neutralizing activity is almost entirely directed to the S1 subunit and that variant cross-neutralization is mediated solely by receptor binding domain-specific antibodies. Our data provide a quantitative framework for guiding future S engineering efforts to develop vaccines with higher resilience to the emergence of variants than current technologies.

Accelerating the Ensemble Convergence of RNA Hairpin Simulations with a Replica Exchange Structure Reservoir.

RNA is a key participant in many biological processes, but studies of RNA using computer simulations lag behind those of proteins, largely due to less-developed force fields and the slow dynamics of RNA. Generating converged RNA ensembles for force field development and other studies remains a challenge. In this study, we explore the ability of replica exchange molecular dynamics to obtain well-converged conformational ensembles for two RNA hairpin systems in an implicit solvent. Even for these small model systems, standard REMD remains computationally costly, but coupling to a pre-generated structure library using the reservoir REMD approach provides a dramatic acceleration of ensemble convergence for both model systems. Such precise ensembles could facilitate RNA force field development and validation and applications of simulation to more complex RNA systems. The advantages and remaining challenges of applying R-REMD to RNA are investigated in detail.

GlycoGrip: Cell Surface-Inspired Universal Sensor for Betacoronaviruses.

Inspired by the role of cell-surface glycoproteins as coreceptors for pathogens, we report the development of GlycoGrip: a glycopolymer-based lateral flow assay for detecting SARS-CoV-2 and its variants. GlycoGrip utilizes glycopolymers for primary capture and antispike antibodies labeled with gold nanoparticles for signal-generating detection. A lock-step integration between experiment and computation has enabled efficient optimization of GlycoGrip test strips which can selectively, sensitively, and rapidly detect SARS-CoV-2 and its variants in biofluids. Employing the power of the glycocalyx in a diagnostic assay has distinct advantages over conventional immunoassays as glycopolymers can bind to antigens in a multivalent capacity and are highly adaptable for mutated strains. As new variants of SARS-CoV-2 are identified, GlycoGrip will serve as a highly reconfigurable biosensor for their detection. Additionally, via extensive ensemble-based docking simulations which incorporate protein and glycan motion, we have elucidated important clues as to how heparan sulfate and other glycocalyx components may bind the spike glycoprotein during SARS-CoV-2 host-cell infection. GlycoGrip is a promising and generalizable alternative to costly, labor-intensive RT-PCR, and we envision it will be broadly useful, including for rural or low-income populations that are historically undertested and under-reported in infection statistics.

Free Energy Landscapes from SARS-CoV-2 Spike Glycoprotein Simulations Suggest that RBD Opening Can Be Modulated via Interactions in an Allosteric Pocket.

The SARS-CoV-2 coronavirus is an enveloped, positive-sense single-stranded RNA virus that is responsible for the COVID-19 pandemic. The spike is a class I viral fusion glycoprotein that extends from the viral surface and is responsible for viral entry into the host cell and is the primary target of neutralizing antibodies. The receptor binding domain (RBD) of the spike samples multiple conformations in a compromise between evading immune recognition and searching for the host-cell surface receptor. Using atomistic simulations of the glycosylated wild-type spike in the closed and 1-up RBD conformations, we map the free energy landscape for RBD opening and identify interactions in an allosteric pocket that influence RBD dynamics. The results provide an explanation for experimental observation of increased antibody binding for a clinical variant with a substitution in this pocket. Our results also suggest the possibility of allosteric targeting of the RBD equilibrium to favor open states via binding of small molecules to the hinge pocket. In addition to potential value as experimental probes to quantify RBD conformational heterogeneity, small molecules that modulate the RBD equilibrium could help explore the relationship between RBD opening and S1 shedding.

Scaffold Hopping Transformations Using Auxiliary Restraints for Calculating Accurate Relative Binding Free Energies.

In silico screening of drug-target interactions is a key part of the drug discovery process. Changes in the drug scaffold via contraction or expansion of rings, the breaking of rings, and the introduction of cyclic structures from acyclic structures are commonly applied by medicinal chemists to improve binding affinity and enhance favorable properties of candidate compounds. These processes, commonly referred to as scaffold hopping, are challenging to model computationally. Although relative binding free energy (RBFE) calculations have shown success in predicting binding affinity changes caused by perturbing R-groups attached to a common scaffold, applications of RBFE calculations to modeling scaffold hopping are relatively limited. Scaffold hopping inevitably involves breaking and forming bond interactions of quadratic functional forms, which is highly challenging. A novel method for handling ring opening/closure/contraction/expansion and linker contraction/expansion is presented here. To the best of our knowledge, RBFE calculations on linker contraction/expansion have not been previously reported. The method uses auxiliary restraints to hold the atoms at the ends of a bond in place during the breaking and forming of the bonds. The broad applicability of the method was demonstrated by examining perturbations involving small-molecule macrocycles and mutations of proline in proteins. High accuracy was obtained using the method for most of the perturbations studied. The rigor of the method was isolated from the force field by validating the method using relative and absolute hydration free energy calculations compared to standard simulation results. Unlike other methods that rely on λ-dependent functional forms for bond interactions, the method presented here can be employed using modern molecular dynamics software without modification of codes or force field functions.

Quantitative Analysis of Protein Unfolded State Energetics: Experimental and Computational Studies Demonstrate That Non-Native Side-Chain Interactions Stabilize Local Native Backbone Structure.

Proteins fold on relatively smooth free energy landscapes which are biased toward the native state, but even simple topologies which fold rapidly can experience roughness on their free energy landscape. The details of these interactions are difficult to elucidate experimentally. Closely related to the problem of deciphering the details of the free energy landscape is the problem of defining the interactions in the denatured state ensemble (DSE) which is populated under native conditions, that is, under conditions where the native state is stable. The DSE of many proteins deviates from random coil models, but quantifying and defining the energetics of the transiently populated interactions in this ensemble is extremely challenging. Characterization of the DSE of proteins which fold to compact structures is also relevant to studies of intrinsically disordered proteins (IDPs) since interactions in the dynamic ensemble populated by IDPs can modulate their behavior. Here we show how experimental thermodynamic and pKa measurements can be combined with computational thermodynamic integration to quantify interactions in the DSE. We show that non-native side chain interactions can stabilize native backbone structure in the DSE and demonstrate that that even rapidly folding proteins can form energetically significant non-native interactions in their DSE. As an example, we characterize a non-native salt bridge that stabilizes local native backbone structure in the DSE of a widely studied fast-folding protein, the villin headpiece helical domain. The combined computational experimental approach is applicable to other protein unfolded states and provides insight that is impossible to obtain with either method alone.

The flexibility of ACE2 in the context of SARS-CoV-2 infection.

The coronavirus disease 2019 (COVID-19) pandemic has swept over the world in the past months, causing significant loss of life and consequences to human health. Although numerous drug and vaccine development efforts are underway, there are many outstanding questions on the mechanism of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) viral association to angiotensin-converting enzyme 2 (ACE2), its main host receptor, and host cell entry. Structural and biophysical studies indicate some degree of flexibility in the viral extracellular spike glycoprotein and at the receptor-binding domain (RBD)-receptor interface, suggesting a role in infection. Here, we perform explicitly solvated, all-atom, molecular dynamics simulations of the glycosylated, full-length, membrane-bound ACE2 receptor in both an apo and spike RBD-bound state to probe the intrinsic dynamics of the ACE2 receptor in the context of the cell surface. A large degree of fluctuation in the full-length structure is observed, indicating hinge bending motions at the linker region connecting the head to the transmembrane helix while still not disrupting the ACE2 homodimer or ACE2-RBD interfaces. This flexibility translates into an ensemble of ACE2 homodimer conformations that could sterically accommodate binding of the spike trimer to more than one ACE2 homodimer and suggests a mechanical contribution of the host receptor toward the large spike conformational changes required for cell fusion. This work presents further structural and functional insights into the role of ACE2 in viral infection that can potentially be exploited for the rational design of effective SARS-CoV-2 therapeutics.

AI-driven multiscale simulations illuminate mechanisms of SARS-CoV-2 spike dynamics.

We develop a generalizable AI-driven workflow that leverages heterogeneous HPC resources to explore the time-dependent dynamics of molecular systems. We use this workflow to investigate the mechanisms of infectivity of the SARS-CoV-2 spike protein, the main viral infection machinery. Our workflow enables more efficient investigation of spike dynamics in a variety of complex environments, including within a complete SARS-CoV-2 viral envelope simulation, which contains 305 million atoms and shows strong scaling on ORNL Summit using NAMD. We present several novel scientific discoveries, including the elucidation of the spike's full glycan shield, the role of spike glycans in modulating the infectivity of the virus, and the characterization of the flexible interactions between the spike and the human ACE2 receptor. We also demonstrate how AI can accelerate conformational sampling across different systems and pave the way for the future application of such methods to additional studies in SARS-CoV-2 and other molecular systems.

AI-Driven Multiscale Simulations Illuminate Mechanisms of SARS-CoV-2 Spike Dynamics.

We develop a generalizable AI-driven workflow that leverages heterogeneous HPC resources to explore the time-dependent dynamics of molecular systems. We use this workflow to investigate the mechanisms of infectivity of the SARS-CoV-2 spike protein, the main viral infection machinery. Our workflow enables more efficient investigation of spike dynamics in a variety of complex environments, including within a complete SARS-CoV-2 viral envelope simulation, which contains 305 million atoms and shows strong scaling on ORNL Summit using NAMD. We present several novel scientific discoveries, including the elucidation of the spike's full glycan shield, the role of spike glycans in modulating the infectivity of the virus, and the characterization of the flexible interactions between the spike and the human ACE2 receptor. We also demonstrate how AI can accelerate conformational sampling across different systems and pave the way for the future application of such methods to additional studies in SARS-CoV-2 and other molecular systems.

Protein storytelling through physics.

Every protein has a story-how it folds, what it binds, its biological actions, and how it misbehaves in aging or disease. Stories are often inferred from a protein's shape (i.e., its structure). But increasingly, stories are told using computational molecular physics (CMP). CMP is rooted in the principled physics of driving forces and reveals granular detail of conformational populations in space and time. Recent advances are accessing longer time scales, larger actions, and blind testing, enabling more of biology's stories to be told in the language of atomistic physics.