Reinvent 4: Modern AI-driven generative molecule design.

REINVENT 4 is a modern open-source generative AI framework for the design of small molecules. The software utilizes recurrent neural networks and transformer architectures to drive molecule generation. These generators are seamlessly embedded within the general machine learning optimization algorithms, transfer learning, reinforcement learning and curriculum learning. REINVENT 4 enables and facilitates de novo design, R-group replacement, library design, linker design, scaffold hopping and molecule optimization. This contribution gives an overview of the software and describes its design. Algorithms and their applications are discussed in detail. REINVENT 4 is a command line tool which reads a user configuration in either TOML or JSON format. The aim of this release is to provide reference implementations for some of the most common algorithms in AI based molecule generation. An additional goal with the release is to create a framework for education and future innovation in AI based molecular design. The software is available from https://github.com/MolecularAI/REINVENT4 and released under the permissive Apache 2.0 license. Scientific contribution. The software provides an open-source reference implementation for generative molecular design where the software is also being used in production to support in-house drug discovery projects. The publication of the most common machine learning algorithms in one code and full documentation thereof will increase transparency of AI and foster innovation, collaboration and education.

Reproducibility of Free Energy Calculations across Different Molecular Simulation Software Packages.

Alchemical free energy calculations are an increasingly important modern simulation technique to calculate free energy changes on binding or solvation. Contemporary molecular simulation software such as AMBER, CHARMM, GROMACS, and SOMD include support for the method. Implementation details vary among those codes, but users expect reliability and reproducibility, i.e., for a given molecular model and set of force field parameters, comparable free energy differences should be obtained within statistical bounds regardless of the code used. Relative alchemical free energy (RAFE) simulation is increasingly used to support molecule discovery projects, yet the reproducibility of the methodology has been less well tested than its absolute counterpart. Here we present RAFE calculations of hydration free energies for a set of small organic molecules and demonstrate that free energies can be reproduced to within about 0.2 kcal/mol with the aforementioned codes. Absolute alchemical free energy simulations have been carried out as a reference. Achieving this level of reproducibility requires considerable attention to detail and package-specific simulation protocols, and no universally applicable protocol emerges. The benchmarks and protocols reported here should be useful for the community to validate new and future versions of software for free energy calculations.

Approaches for calculating solvation free energies and enthalpies demonstrated with an update of the FreeSolv database.

Solvation free energies can now be calculated precisely from molecular simulations, providing a valuable test of the energy functions underlying these simulations. Here, we briefly review "alchemical" approaches for calculating the solvation free energies of small, neutral organic molecules from molecular simulations, and illustrate by applying them to calculate aqueous solvation free energies (hydration free energies). These approaches use a non-physical pathway to compute free energy differences from a simulation or set of simulations and appear to be a particularly robust and general-purpose approach for this task. We also present an update (version 0.5) to our FreeSolv database of experimental and calculated hydration free energies of neutral compounds and provide input files in formats for several simulation packages. This revision to FreeSolv provides calculated values generated with a single protocol and software version, rather than the heterogeneous protocols used in the prior version of the database. We also further update the database to provide calculated enthalpies and entropies of hydration and some experimental enthalpies and entropies, as well as electrostatic and nonpolar components of solvation free energies.

FESetup: Automating Setup for Alchemical Free Energy Simulations.

FESetup is a new pipeline tool which can be used flexibly within larger workflows. The tool aims to support fast and easy setup of alchemical free energy simulations for molecular simulation packages such as AMBER, GROMACS, Sire, or NAMD. Post-processing methods like MM-PBSA and LIE can be set up as well. Ligands are automatically parametrized with AM1-BCC, and atom mappings for a single topology description are computed with a maximum common substructure search (MCSS) algorithm. An abstract molecular dynamics (MD) engine can be used for equilibration prior to free energy setup or standalone. Currently, all modern AMBER force fields are supported. Ease of use, robustness of the code, and automation where it is feasible are the main development goals. The project follows an open development model, and we welcome contributions.

Evaluation of Selected Classical Force Fields for Alchemical Binding Free Energy Calculations of Protein-Carbohydrate Complexes.

Protein-carbohydrate recognition is crucial in many vital biological processes including host-pathogen recognition, cell-signaling, and catalysis. Accordingly, computational prediction of protein-carbohydrate binding free energies is of enormous interest for drug design. However, the accuracy of current force fields (FFs) for predicting binding free energies of protein-carbohydrate complexes is not well understood owing to technical challenges such as the highly polar nature of the complexes, anomerization, and conformational flexibility of carbohydrates. The present study evaluated the performance of alchemical predictions of binding free energies with the GAFF1.7/AM1-BCC and GLYCAM06j force fields for modeling protein-carbohydrate complexes. Mean unsigned errors of 1.1 ± 0.06 (GLYCAM06j) and 2.6 ± 0.08 (GAFF1.7/AM1-BCC) kcal·mol(-1) are achieved for a large data set of monosaccharide ligands for Ralstonia solanacearum lectin (RSL). The level of accuracy provided by GLYCAM06j is sufficient to discriminate potent, moderate, and weak binders, a goal that has been difficult to achieve through other scoring approaches. Accordingly, the protocols presented here could find useful applications in carbohydrate-based drug and vaccine developments.

Ligand binding and dynamics of the monomeric epidermal growth factor receptor ectodomain.

The ectodomain of the human epidermal growth factor receptor (hEGFR) controls input to several cell signalling networks via binding with extracellular growth factors. To gain insight into the dynamics and ligand binding of the ectodomain, the hEGFR monomer was subjected to molecular dynamics simulation. The monomer was found to be substantially more flexible than the ectodomain dimer studied previously. Simulations where the endogeneous ligand EGF binds to either Subdomain I or Subdomain III, or where hEGFR is unbound, show significant differences in dynamics. The molecular mechanics Poisson-Boltzmann surface area method has been used to derive relative free energies of ligand binding, and we find that the ligand is capable of binding either subdomain with a slight preference for III. Alanine-scanning calculations for the effect of selected ligand mutants on binding reproduce the trends of affinity measurements. Taken together, these results emphasize the possible role of the ectodomain monomer in the initial step of ligand binding, and add details to the static picture obtained from crystal structures.

Human epidermal growth factor receptor (EGFR) aligned on the plasma membrane adopts key features of Drosophila EGFR asymmetry.

The ability of epidermal growth factor receptor (EGFR) to control cell fate is defined by its affinity for ligand. Current models suggest that ligand-binding heterogeneity arises from negative cooperativity in signaling receptor dimers, for which the asymmetry of the extracellular region of the Drosophila EGFR has recently provided a structural basis. However, no asymmetry is apparent in the isolated extracellular region of the human EGFR. Human EGFR also differs from the Drosophila EGFR in that negative cooperativity is found only in full-length receptors in cells. To gain structural insights into the human EGFR in situ, we developed an approach based on quantitative Förster resonance energy transfer (FRET) imaging, combined with Monte Carlo and molecular dynamics simulations, to probe receptor conformation in epithelial cells. We experimentally demonstrate a high-affinity ligand-binding human EGFR conformation consistent with the extracellular region aligned flat on the plasma membrane. We explored the relevance of this conformation to ligand-binding heterogeneity and found that the asymmetry of this structure shares key features with that of the Drosophila EGFR, suggesting that the structural basis for negative cooperativity is conserved from invertebrates to humans but that in human EGFR the extracellular region asymmetry requires interactions with the plasma membrane.

Collective dynamics of periplasmic glutamine binding protein upon domain closure.

The glutamine binding protein is a vital component of the associated ATP binding cassette transport systems responsible for the uptake of glutamine into the cell. We have investigated the global movements of this protein by molecular dynamics simulations and principal component analysis (PCA). We confirm that the most dominant mode corresponds to the biological function of the protein, i.e., a hinge-type motion upon ligand binding. The closure itself was directly observed from two independent trajectories whereby PCA was used to elucidate the nature of this closing reaction. Two intermediary states are identified and described in detail. The ligand binding induces the structural change of the hinge regions from a discontinuous beta-sheet to a continuous one, which also enhances softness of the hinge and modifies the direction of hinge motion to enable closing. We also investigated the convergence behavior of PCA modes, which were found to converge rather quickly when the associated magnitudes of the eigenvalues are well separated.

Ectodomain orientation, conformational plasticity and oligomerization of ErbB1 receptors investigated by molecular dynamics.

Epidermal growth factor receptor (EGFR; ErbB1, HER1 in humans) is a receptor tyrosine kinase triggering signals across the plasma membranes of cells to determine cell fate. We have used molecular dynamics simulations to investigate structural models of ErbB1 ectodomains. We show that, with minor rearrangements, the ErbB1 back-to-back dimer can align almost flat on the cell membrane. This is in contrast to the traditional picture of ErbB1 dimers standing proud of the membrane, but in line with recent FRET and EM experiments. Interaction with the membrane leads to conformational changes in the dimer, which further stabilize the back-to-back interface. On the membrane, two dimers can associate forming a tetramer. This is enabled by a head-to-head interface, involving the ligand binding side of the ectodomain, and which significantly enhances ligand binding. A weak head-to-head interface has been seen in crystal structures, but is found to stabilise appreciably in our simulation. We also find that the domains IV, connecting the receptor to the membrane, weakly interact with each other. These simulations illustrate some of the flexibility of the ErbB1 ectodomains, and may help to explain recent experimental results.

Water exchange dynamics of lithium(I) ion in aqueous solution.

The water exchange dynamics of the fourfold coordinated first hydration shell of the lithium(I) ion was studied by both direct and umbrella sampling QM/MM-MD and classical MD simulations. The structural changes and energetics accompanying the activation process are discussed. The overall exchange rate constant was found to be k(ex) = 5.8 x 10(9) s(-1) from classical MD simulations. QM/MM-MD umbrella sampling simulations predict an exchange rate constant of k(TST) = 1.01 x 10(10) s(-1) as obtained from classical transition-state theory. First-shell ligands exchange preferably via an associatively activated mode.

Influence of backbone conformations of human carbonic anhydrase II on carbon dioxide hydration: hydration pathways and binding of bicarbonate.

In this study, the hydration of carbon dioxide and the formation of bicarbonate in human carbonic anhydrase II have been examined. From semiempirical QM/MM molecular dynamics studies, dominant conformations of the protein backbone, possibly contributing to the catalytic activity, have been isolated and further examined by means of density functional QM/MM methods. In agreement with experimental observations, a binding site for cyanate, which acts as an inhibitor, has been located, whereas for carbon dioxide, depending on the conformation of the protein environment, either a different binding site or no binding site has been found. In the latter case, carbon dioxide diffuses barrierless to the zinc-bound oxygen, and then a weakly bound bicarbonate complex is formed. The formed complex is characterized by a long C-O bond to the zinc-bound hydroxide. The nature of the calculated stationary points was verified by determination of vibrational frequencies. Finally, the dissociation of the formed bicarbonate from zinc has been considered. Therefore, a water molecule was included in the QM zone of the QM/MM hybrid potential, and minimization yielded a pentacoordinated intermediate. From a potential energy scan, an activation energy of 6.2 kcal/mol for dissociation of bicarbonate from Zn has been found.

Many-body effects on structure and dynamics of aqueous ionic solutions.

We performed several molecular dynamic studies of metal cations in aqueous solution. The alkali metal ion Li(+) and the first-row transition metal ion Mn(2+) have been chosen as model systems. Two different three-body corrections are proposed to mimic the crucial many-body effects of electrolyte solutions. The correction function, which includes attractive features of the three-body potential, performs considerably better than the purely repulsive interaction function. Structural and dynamic results show that this simple enhancement is able to satisfactorily reproduce experimental and higher-level results for the first hydration shell.

A QM-MM interface between CHARMM and TURBOMOLE: implementation and application to systems in bulk phase and biologically active systems.

The implementation of a hybrid QM-MM approach combining ab initio and density functional methods of TURBOMOLE with the molecular mechanics program package CHARMM is described. An interface has been created to allow data exchange between the two applications. With this method the efficient multiprocessor capabilities of TURBOMOLE can be utilized with CHARMM running as a single processor application. Therefore, features of nonparallel running code in CHARMM like the TRAVEL module for locating saddle points or VIBRAN for the calculation of second derivatives can be exploited by running the CPU intensive QM calculations in parallel. To test the methodology, several small systems are studied with both Hartree-Fock and density functional methods and varying QM-MM boundaries. Also, the computationally efficient RI-J method has been examined for use in QM-MM applications. A B(12) cofactor containing cobalt has been studied, to examine systems with a large QM region and transition metals. All tested methods perform satisfactory in comparison with pure quantum calculations. Additionally, algorithms for the characterization of saddle points have been tested for their potential use in QM-MM problems. The TRAVEL module of CHARMM has been applied to the Menshutkin reaction in the condensed phase, and a saddle point was located. This saddle point was verified by calculation of a steepest descent path connecting educt, transition state, and product, and by calculation of vibrational modes.

Structure and dynamics of metal ions in solution: QM/MM molecular dynamics simulations of Mn(2+) and V(2+).

Structural and dynamical properties of the transition metal ions V(2+) and Mn(2+) in aqueous solution, resulting from combined quantum mechanical (QM)/molecular mechanical (MM) molecular dynamics (MD) simulations have been compared. The necessity of polarization functions on the ligand's oxygen for a satisfactory description of such ions in aqueous solution is shown using V(2+) as test case. Radial distribution functions, coordination number distributions, and several angle distributions were pursued for a detailed structural comparison of the first hydration shells. Dynamical properties, such as the librational and vibrational motions of water molecules were evaluated by means of velocity autocorrelation functions. Approximative normal coordinate analyses were employed to calculate the rotational frequencies and vibrational motions around the three principal axes. The very low exchange rates for the first shell water exchanges only allow an investigation of the water exchange processes in the second shell, which take place within the picosecond range.