CodonBert large language model for mRNA vaccines.

mRNA-based vaccines and therapeutics are gaining popularity and usage across a wide range of conditions. One of the critical issues when designing such mRNAs is sequence optimization. Even small proteins or peptides can be encoded by an enormously large number of mRNAs. The actual mRNA sequence can have a large impact on several properties including expression, stability, immunogenicity, and more. To enable the selection of an optimal sequence, we developed CodonBERT, a large language model (LLM) for mRNAs. Unlike prior models, CodonBERT uses codons as inputs which enables it to learn better representations. CodonBERT was trained using more than 10 million mRNA sequences from a diverse set of organisms. The resulting model captures important biological concepts. CodonBERT can also be extended to perform prediction tasks for various mRNA properties. CodonBERT outperforms previous mRNA prediction methods including on a new flu vaccine dataset.

Representations of lipid nanoparticles using large language models for transfection efficiency prediction.

MOTIVATION: Lipid nanoparticles (LNPs) are the most widely used vehicles for mRNA vaccine delivery. The structure of the lipids composing the LNPs can have a major impact on the effectiveness of the mRNA payload. Several properties should be optimized to improve delivery and expression including biodegradability, synthetic accessibility, and transfection efficiency. RESULTS: To optimize LNPs, we developed and tested models that enable the virtual screening of LNPs with high transfection efficiency. Our best method uses the lipid Simplified Molecular-Input Line-Entry System (SMILES) as inputs to a large language model. Large language model-generated embeddings are then used by a downstream gradient-boosting classifier. As we show, our method can more accurately predict lipid properties, which could lead to higher efficiency and reduced experimental time and costs. AVAILABILITY AND IMPLEMENTATION: Code and data links available at: https://github.com/Sanofi-Public/LipoBART.

Surface ID: a geometry-aware system for protein molecular surface comparison.

MOTIVATION: A protein can be represented in several forms, including its 1D sequence, 3D atom coordinates, and molecular surface. A protein surface contains rich structural and chemical features directly related to the protein's function such as its ability to interact with other molecules. While many methods have been developed for comparing the similarity of proteins using the sequence and structural representations, computational methods based on molecular surface representation are limited. RESULTS: Here, we describe "Surface ID," a geometric deep learning system for high-throughput surface comparison based on geometric and chemical features. Surface ID offers a novel grouping and alignment algorithm useful for clustering proteins by function, visualization, and in silico screening of potential binding partners to a target molecule. Our method demonstrates top performance in surface similarity assessment, indicating great potential for protein functional annotation, a major need in protein engineering and therapeutic design. AVAILABILITY AND IMPLEMENTATION: Source code for the Surface ID model, trained weights, and inference script are available at https://github.com/Sanofi-Public/LMR-SurfaceID.

Heterogeneous Interactions of Prevalent Indoor Oxygenated Organic Compounds on Hydroxylated SiO2 Surfaces.

Oxygenated organic compounds (OOCs) are widely found in indoor environments and come from either the direct emissions from indoor activities or the subsequent oxidation of nonoxygenated OCs. Adsorption and partitioning of OCs on surfaces are significant processes in indoor chemistry, yet these interactions specifically involving OOCs are still poorly understood. In this study, we investigate the interactions of three prevalent indoor OOCs (dihydromyrcenol, α-terpineol, and linalool) on an indoor surface proxy (hydroxylated SiO2) by combining vibrational spectroscopy with ab initio molecular dynamics simulations. The adsorption of these compounds on the SiO2 surface is driven by π hydrogen bonding and O-H hydrogen bonding interactions, with O-H hydrogen bonding interactions being stronger. The results of kinetic measurements suggest that indoor surfaces play a significant role in the removal of these OOCs, especially under moderate and low air exchange. Additionally, indoor surfaces can also serve as a reservoir of OOCs due to their much slower desorption kinetics when compared to other indoor relevant organic compounds such as limonene. Overall, the results gleaned by experiment and theoretical simulations provide a molecular representation of the interaction of OOCs on indoor relevant surfaces as well as implications of these interactions for indoor air chemistry.

Adsorption of constitutional isomers of cyclic monoterpenes on hydroxylated silica surfaces.

We present a study of four monoterpene isomers (limonene, γ-terpinene, terpinolene, and α-pinene) that are prevalent in indoor environments and their interaction with the hydroxylated SiO2 surface, a model for the glass surface, by combining infrared spectroscopy and computational simulations. These isomers are molecularly adsorbed onto SiO2 through π-hydrogen bonds with surface hydroxyl groups. However, experimental results suggest that the strength of interaction of these compounds with the SiO2 surface varies for each isomer, with α-pinene showing the weakest interaction. This observation is supported by molecular dynamics simulations that α-pinene adsorbed on the SiO2 surface has lower free energy of desorption and a lower mass accommodation coefficient compared to other isomers. Additionally, our ab initio molecular dynamics simulations show lower π-hydrogen bonding probabilities for α-pinene compared to the other three constitutional isomers. Importantly, these interactions are most likely present for a range of other systems involving organic compounds and solid surfaces and, thus, provide a thorough framework for comparing the interactions of organic molecules on indoor relevant surfaces.

Impact of Adsorbed Water on the Interaction of Limonene with Hydroxylated SiO2: Implications of π-Hydrogen Bonding for Surfaces in Humid Environments.

The indoor environment is a dynamic one with many variables impacting indoor air quality and indoor air chemistry. These include relative humidity (RH) and the presence of different surfaces. Although it has been suggested that the indoor concentrations of gas-phase compounds increase at higher relative humidity, because of displacement of these compounds from indoor surfaces, little is known from a molecular perspective about how RH and adsorbed water impact the adsorption of indoor relevant organic compounds such as limonene with indoor relevant surfaces. Herein, we investigate the effects of RH on the adsorption of limonene, a hydrophobic molecule, on hydroxylated SiO2 surfaces, a model for glass surfaces. Experimental data using infrared spectroscopy to directly measure limonene adsorption are combined with both force field-based molecular dynamics (MD) and ab initio molecular dynamics (AIMD) simulations to understand the competitive interactions between limonene, water, and the SiO2 surface. The spectroscopic data provide evidence that adsorbed limonene is not completely displaced by adsorbed water, even at high RH (∼80%) when the water layer coverage is close to three monolayers (MLs). These experimental data are supported by AIMD and MD simulations, which indicate that limonene is present at the adsorbed water interface but displaced from direct interactions with SiO2. This study shows that although some limonene can desorb from the surface, even at the highest RH, more than half the limonene remains adsorbed on the surface that can undergo continued surface reactivity. A complex network of π-hydrogen bonds, water-water hydrogen bonds, and SiO2-water hydrogen bonds explains these interactions at the air/adsorbed water/SiO2 interface that hold the hydrophobic limonene molecule at the interface. Importantly, these interactions are most likely present for a range of other systems involving organic compounds and solid surfaces at ambient relative humidity and may be important in a range of scientific areas, from sensor development to cultural heritage science.

Why can hydrogen sulfide permeate cell membranes?

The high membrane permeability of H2S was studied using polarizable molecular dynamics simulations of a DPPC lipid bilayer. The solubility-diffusion model predicts permeability coefficients of H2S and H2O that are in good agreement with experiment. The computed diffusion coefficient profile shows H2S to diffuse at a lower rate than H2O, but the barrier for H2S permeation on the Gibbs energy profile is negligible. The hydrophobicity of H2S allows it to partition into the paraffinic interior of the membrane readily.

The CHARMM-TURBOMOLE interface for efficient and accurate QM/MM molecular dynamics, free energies, and excited state properties.

The quantum mechanical (QM)/molecular mechanical (MM) interface between Chemistry at HARvard Molecular Mechanics (CHARMM) and TURBOMOLE is described. CHARMM provides an extensive set of simulation algorithms, like molecular dynamics (MD) and free energy perturbation, and support for mature nonpolarizable and Drude polarizable force fields. TURBOMOLE provides fast QM calculations using density functional theory or wave function methods and excited state properties. CHARMM-TURBOMOLE is well-suited for extended QM/MM MD simulations using first principles methods with large (triple-ζ) basis sets. We demonstrate these capabilities with a QM/MM simulation of Mg(2+) (aq), where the MM outer sphere water molecules are represented using the SWM4-NDP Drude polarizable force field and the ion and inner coordination sphere are represented using QM PBE, PBE0, and MP2 methods. The relative solvation free energies of Mg(2+) and Zn(2+) were calculated using thermodynamic integration. We also demonstrate the features for excited state properties. We calculate the time-averaged solution absorption spectrum of indole, the emission spectrum of the indole 1La excited state, and the electronic circular dichroism spectrum of an oxacepham.

A novel Hv1 inhibitor reveals a new mechanism of inhibition of a voltage-sensing domain.

Voltage-gated sodium, potassium, and calcium channels consist of four voltage-sensing domains (VSDs) that surround a central pore domain and transition from a down state to an up state in response to membrane depolarization. While many types of drugs bind pore domains, the number of organic molecules known to bind VSDs is limited. The Hv1 voltage-gated proton channel is made of two VSDs and does not contain a pore domain, providing a simplified model for studying how small ligands interact with VSDs. Here, we describe a ligand, named HIF, that interacts with the Hv1 VSD in the up and down states. We find that HIF rapidly inhibits proton conduction in the up state by blocking the open channel, as previously described for 2-guanidinobenzimidazole and its derivatives. HIF, however, interacts with a site slowly accessible in the down state. Functional studies and MD simulations suggest that this interaction traps the compound in a narrow pocket lined with charged residues within the VSD intracellular vestibule, which results in slow recovery from inhibition. Our findings point to a "wrench in gears" mechanism whereby side chains within the binding pocket trap the compound as the teeth of interlocking gears. We propose that the use of screening strategies designed to target binding sites with slow accessibility, similar to the one identified here, could lead to the discovery of new ligands capable of interacting with VSDs of other voltage-gated ion channels in the down state.

HIFs: New arginine mimic inhibitors of the Hv1 channel with improved VSD-ligand interactions.

The human voltage-gated proton channel Hv1 is a drug target for cancer, ischemic stroke, and neuroinflammation. It resides on the plasma membrane and endocytic compartments of a variety of cell types, where it mediates outward proton movement and regulates the activity of NOX enzymes. Its voltage-sensing domain (VSD) contains a gated and proton-selective conduction pathway, which can be blocked by aromatic guanidine derivatives such as 2-guanidinobenzimidazole (2GBI). Mutation of Hv1 residue F150 to alanine (F150A) was previously found to increase 2GBI apparent binding affinity more than two orders of magnitude. Here, we explore the contribution of aromatic interactions between the inhibitor and the channel in the presence and absence of the F150A mutation, using a combination of electrophysiological recordings, classic mutagenesis, and site-specific incorporation of fluorinated phenylalanines via nonsense suppression methodology. Our data suggest that the increase in apparent binding affinity is due to a rearrangement of the binding site allowed by the smaller residue at position 150. We used this information to design new arginine mimics with improved affinity for the nonrearranged binding site of the wild-type channel. The new compounds, named "Hv1 Inhibitor Flexibles" (HIFs), consist of two "prongs," an aminoimidazole ring, and an aromatic group connected by extended flexible linkers. Some HIF compounds display inhibitory properties that are superior to those of 2GBI, thus providing a promising scaffold for further development of high-affinity Hv1 inhibitors.

Application of an integrated computational antibody engineering platform to design SARS-CoV-2 neutralizers.

As the COVID-19 pandemic continues to spread, hundreds of new initiatives including studies on existing medicines are running to fight the disease. To deliver a potentially immediate and lasting treatment to current and emerging SARS-CoV-2 variants, new collaborations and ways of sharing are required to create as many paths forward as possible. Here, we leverage our expertise in computational antibody engineering to rationally design/engineer three previously reported SARS-CoV neutralizing antibodies and share our proposal towards anti-SARS-CoV-2 biologics therapeutics. SARS-CoV neutralizing antibodies, m396, 80R and CR-3022 were chosen as templates due to their diversified epitopes and confirmed neutralization potency against SARS-CoV (but not SARS-CoV-2 except for CR3022). Structures of variable fragment (Fv) in complex with receptor binding domain (RBD) from SARS-CoV or SARS-CoV-2 were subjected to our established in silico antibody engineering platform to improve their binding affinity to SARS-CoV-2 and developability profiles. The selected top mutations were ensembled into a focused library for each antibody for further screening. In addition, we convert the selected binders with different epitopes into the trispecific format, aiming to increase potency and to prevent mutational escape. Lastly, to avoid antibody-induced virus activation or enhancement, we suggest application of NNAS and DQ mutations to the Fc region to eliminate effector functions and extend half-life.

Embedded Mean-Field Theory for Solution-Phase Transition-Metal Polyolefin Catalysis.

Decreasing the wall-clock time of quantum mechanics/molecular mechanics (QM/MM) calculations without sacrificing accuracy is a crucial prerequisite for widespread simulation of solution-phase dynamical processes. In this work, we demonstrate the use of embedded mean-field theory (EMFT) as the QM engine in QM/MM molecular dynamics (MD) simulations to examine polyolefin catalysts in solution. We show that employing EMFT in this mode preserves the accuracy of hybrid-functional DFT in the QM region, while providing up to 20-fold reductions in the cost per SCF cycle, thereby increasing the accessible simulation time-scales. We find that EMFT reproduces DFT-computed binding energies and optimized bond lengths to within chemical accuracy, as well as consistently ranking conformer stability. Furthermore, solution-phase EMFT/MM simulations provide insight into the interaction strength of strongly coordinating and bulky counterions.

What Is the Driving Force behind the Adsorption of Hydrophobic Molecules on Hydrophilic Surfaces?

The adsorption of limonene, a common organic compound found in indoor air, on hydrophilic surfaces such as glass (SiO2), a prevalent surface in the indoor environment, is poorly understood. In this study, we have investigated the interaction of limonene and three other cyclic hydrocarbons (cyclohexane, cyclohexene, and benzene) on hydroxylated SiO2 using infrared spectroscopy and ab initio molecular dynamics (AIMD) simulations. Experimental results show that there is an interaction between these cyclic hydrocarbons and surface hydroxyl groups. AIMD simulations demonstrate that all of the cyclic molecules, except for cyclohexane, π-hydrogen bond with surface hydroxyl groups while cyclohexane interacts with the surface OH groups through dispersion forces. According to experiments and simulations, the intermolecular interaction between limonene and SiO2 is significantly stronger than those of other compounds explored. This study provides an understanding of some of the driving forces behind the formation of organic coatings on glass surfaces important in indoor environments.

A molecular picture of surface interactions of organic compounds on prevalent indoor surfaces: limonene adsorption on SiO2.

Indoor surfaces are often coated with organic compounds yet a molecular understanding of what drives these interactions is poorly understood. Herein, the adsorption and desorption of limonene, an organic compound found in indoor environments, on hydroxylated silica (SiO2) surfaces, used to mimic indoor glass surfaces, is investigated by combining vibrational spectroscopy, atomistic computer simulations and kinetic modeling. Infrared spectroscopy shows the interaction involves hydrogen-bonding between limonene and surface O-H groups. Atomistic molecular dynamics (MD) simulations confirm the existence of π-hydrogen bonding interactions, with one or two hydrogen bonds between the silica O-H groups and the carbon-carbon double bonds, roughly one third of the time. The concentration and temperature dependent adsorption/desorption kinetics as measured by infrared spectroscopy were reproduced with a kinetic model, yielding the adsorption enthalpy of ∼55 kJ mol-1, which is consistent with the value derived from the MD simulations. Importantly, this integrated experimental, theoretical and kinetic modeling study constitutes a conceptual framework for understanding the interaction of organic compounds with indoor relevant surfaces and thus provides important insights into our understanding of indoor air chemistry and indoor air quality.

Solvation of hydrogen sulfide in liquid water and at the water-vapor interface using a polarizable force field.

Molecular dynamics (MD) simulations using the Drude polarizable force field are used to study the solution and interfacial properties of hydrogen sulfide (H2S) in water. Pairwise H2O-H2S Lennard-Jones interactions were optimized to the experimental H2S gas solubility at 298 K. These parameters yield hydration free energies and diffusion coefficients for H2S that are in good agreement with the experiment over 273-323 K and 298-368 K, respectively. H2S is sparingly soluble in water, with a ΔG(hydr)° of -0.5 kcal mol(-1). The free energy perturbation (FEP) calculations and analysis of the radial distribution functions show that H2S has limited hydrogen bonding and electrostatic interactions with the water solvent and generally behaves like a hydrophobic solute. These features were confirmed by ab initio MD simulations. Umbrella sampling simulations were used to calculate the free energy profile of the transition of H2S across the water-vapor interface, which showed that H2S has a sizable surface excess, with a ΔG(surf) of 1.3 kcal mol(-1). This high surface excess is consistent with our calculations of the surface tension, which decreases to 20 dyn cm(-1) under high densities of H2S (g). The dipole moment of H2S increases from its gas phase value of 0.98 to 1.25 D in bulk water as it moves across the interface. Adsorbed H2S tends to be oriented perpendicular to the interface, with the sulfur atom pointing toward the vapor phase.

A Drude polarizable model for liquid hydrogen sulfide.

A polarizable force field for liquid hydrogen sulfide (H2S) has been developed based on the Drude oscillator model. This force field has been designed to be analogous to the SWM4-NDP water model; the model is rigid with point charges assigned to the H and S atoms and a lone pair on the bisector of ∠HSH in the molecular plane. Positions of the lone pair and the charges have been defined such that the model has a static dipole moment of 0.98 D, equal to the experimental value. Polarizability is incorporated by a charged (Drude) particle attached to the S atom through a harmonic potential. Intermolecular nonbonded forces are included by use of a Lennard-Jones potential between S atoms. The model was parametrized to reproduce the density, enthalpy of vaporization, and dielectric constant of pure H2S at 212 K and 1 atm. The calculated density, enthalpy of vaporization, shear viscosity coefficient, and self-diffusion coefficient are in good agreement with experiment over the temperature range 212-298 K along the liquid-vapor coexistence curve of liquid H2S. The radial distribution function calculated from this model is in good agreement with experimental diffraction data and ab initio molecular dynamics simulations.