The optimisation of a peptide-capped glycine using the novel force field FFLUX is presented. FFLUX is a force field based on the machine-learning method kriging and the topological energy partitioning method called Interacting Quantum Atoms. FFLUX has a completely different architecture to that of traditional force fields, avoiding (harmonic) potentials for bonded, valence and torsion angles. In this study, FFLUX performs an optimisation on a glycine molecule and successfully recovers the target density-functionaltheory energy with an error of 0.89 ± 0.03 kJ mol −1. It also recovers the structure of the global minimum with a root-mean-squared deviation of 0.05 Å (excluding hydrogen atoms). We also show that the geometry of the intra-molecular hydrogen bond in glycine is recovered accurately.

FFLUX is a new force field that combines the accuracy of quantum mechanics with the speed of force fields, without any link to the architecture of classical force fields. This force field is atom-focused and adopts the parameter-free topological atom from Quantum Chemical Topology (QCT). FFLUX uses Gaussian Process Regression (GPR) (aka kriging) models to make predicƟons of atomic properties, which in this work are atomic energies according to QCT's InteracƟng Quantum Atom (IQA) approach. Here we report the adaptive sampling technique Maximum Expected Prediction Error (MEPE) to create data-compact, efficient and accurate kriging models (sub kJ mol-1 for water, ammonia, methane and methanol, and sub kcal mol-1 for N-methylacetamide (NMA)). The models cope with large molecular distortions and are ready for use in molecular simulation. A brand new press-one-buƩon Python pipeline, called ICHOR, carries out the training. File list (2) download file view on ChemRxiv AdaptiveSampling_ms_submit.pdf (914.56 KiB) download file view on ChemRxiv AdaptiveSampling_SI_submit.pdf (868.68 KiB)

Molecular details for the timing and role of proton transfer in phosphoryl transfer reactions are poorly understood. Here, we have combined QM models, experimental NMR measurements, and X-ray structures to establish that the transition of an archetypal phosphoryl transfer enzyme, βPGM, from a very closed near-attack conformation to a fully closed transition state analogue (TSA) conformation triggers both partial proton transfer from the general acid–base residue to the leaving group oxygen and partial dissociation of the transferring phosphoryl group from the leaving group oxygen. Proton transfer continues but is not completed throughout the reaction path of the phosphoryl transfer with the enzyme in the TSA conformation. Moreover, using interacting quantum atoms (IQA) and relative energy gradient (REG) analysis approaches, we observed that the change in the position of the proton and the corresponding increased electrostatic repulsion between the proton and the phosphorus atom provide a stimulus for phosphoryl transfer in tandem with a reduction in the negative charge density on the leaving group oxygen atom. The agreement between solution-phase 19F NMR measurements and equivalent QM models of βPGMWT and βPGMD10N TSA complexes confirms the protonation state of G6P in the two variants, validating the employed QM models. Furthermore, QM model predictions of an AlF4 distortion in response to the proton position are confirmed using high resolution X-ray crystal structures, not only providing additional validation to the QM models but also further establishing metal fluorides as highly sensitive experimental predictors of active-site charge density distributions.

<div>FFLUX is a new force field that combines the accuracy of quantum mechanics with the speed of force </div><div>fields, without any link to the architecture of classical force fields. This force field is atom‐focused and </div><div>adopts the parameter‐free topological atom from Quantum Chemical Topology (QCT). FFLUX uses </div><div>Gaussian Process Regression (GPR) (aka kriging) models to make predicƟons of atomic properties, which </div><div>in this work are atomic energies according to QCT’s InteracƟng Quantum Atom (IQA) approach. Here </div><div>we report the adaptive sampling technique Maximum Expected Prediction Error (MEPE) to create data‐</div><div>compact, efficient and accurate kriging models (sub kJ mol‐1 for water, ammonia, methane and </div><div>methanol, and sub kcal mol‐1 for N‐methylacetamide (NMA)). The models cope with large molecular </div><div>distortions and are ready for use in molecular simulation. A brand new press‐one‐buƩon Python </div><div>pipeline, called ICHOR, carries out the training. </div>

FEREBUS is a highly optimised Gaussian process regression (GPR) engine, which provides both model and optimiser flexibility to produce tailored models designed for domain specific applications.