Michael R. Shirts

For many existing and emerging applications, there is a great need to develop novel polymeric materials with highly tuned properties. In the quest to design materials from highly abundant natural resources or waste materials, there is a vast design space that has remained substantially unexplored. Machine learning approaches can help guide toward improved properties, but these approaches often suffer from a lack of high quality data, the acquisition of which can be tedious, time-consuming, and expensive. In silico atomistic simulations are a promising approach to generate data that are highly correlated to real polymer properties, but simulations of polymers present myriad challenges including accurate and automatic force field parameterization, initial configuration generation, and long relaxation times which make it extremely difficult to perform such simulations in a high-throughput manner. Here, we present a computational pipeline starting from SMILES strings of reactants to automatically generate force field parameters and initial system coordinates for subsequent molecular dynamics (MD) simulations of short oligomers that have much faster relaxation times than the corresponding polymer melts. Because most measurable properties of oligomeric fluids are not the same as those of the corresponding polymers of much higher degree of polymerization, and indeed many properties of polymer phases do not exist for such liquids, we employ a transfer learning scheme to predict polymer properties from the oligomer thermophysical data. Using this approach, we are able to achieve improved property predictions for a number of polymer properties, including gas permeabilities. These improvements are most dramatic when experimental data are sparse and when the experimental observables are most directly correlated to simulation properties.

The primary limitation for the application of alchemical free energy methods to a wider variety of complex molecular systems is achieving reasonable sampling. Flexible binding complexes often have high free energy barriers, which require prohibitively long simulations or carefully tuned enhanced sampling methods in order to gather sufficient uncorrelated samples to obtain reliable free energy estimates. An example of such a flexible system is the complex formed between FabB, an elongating β-ketoacyl–acyl carrier protein (ACP) synthase (KS) from Escherichia coli, and ACP, which carries acyl chains of varying lengths. Previous experimental evidence suggests that growing acyl chains can bind to at least two pockets in FabB. With the multiple topology replica exchange of expanded ensemble (MT-REXEE) enhanced sampling approach, we can obtain highly efficient sampling of both pockets by adaptively growing and shrinking the chains in the simulation ensemble, allowing each simulation to visit chain lengths that can transition between pockets. This approach enables unbiased sampling of alternate configurational states for large complex systems without prior pocket definitions, as collective-variable based enhanced sampling methods would require. Using the new swapping approach gives significantly enhanced sampling even for comparatively simple small molecule binding , as demonstrated by faster convergence of free energy estimates of relative binding affinity between mouse major urinary protein 1 ligands. This case study demonstrates the utility of MT-REXEE and its open-source implementation particularly for systems that feature high free energy barriers for a subset of ligands of interest, demonstrating a valuable addition to the existing stable of enhanced sampling methods.

Protein tyrosine kinases (PTKs) regulate cellular biochemistry by phosphorylating tyrosine residues that alter protein function; their substrate preferences define the topology of signaling cascades. Previous studies of PTKs have mapped their average preferences for amino acids surrounding phosphorylation sites, but their sensitivity to sequence variation remains poorly understood. Here, we used microbial biosensors for PTK activity to examine the influence of local sequence context on substrate specificity. Across five well-studied PTKs, we identified amino acid substitutions within consensus substrates that could confer sensitivity to substrate length or enhance selectivity for one PTK over others. Using a secondary decoy screen, we found sequence-diverse substrates with unexpectedly orthogonal PTK compatibilities. Our findings show how context-specific sequence features alter PTK substrate specificity far beyond what might be expected from classical consensus models and establish an experimental framework for defining the limits of substrate overlap between closely related kinases.

Partial atomic charges are a fundamental component underlying classical molecular simulations, but assigning charges remains a computational bottleneck; many common methods rely on quantum mechanical calculations that scale poorly with molecular size and are sensitive to the choice of conformer geometry. We introduce Open Force Field (OpenFF) AshGC, a new graph convolutional neural network charge model, as well as the Sage 2.3.0 small molecule force field for drug-like molecules parametrized to be consistent with AshGC. AshGC is designed to efficiently produce conformer-independent charges of semiempirical quality at linear cost for molecules of all sizes, from small molecules to macromolecules. AshGC largely generates charges within the range of other accepted AM1-BCC backends such as OpenEye's oequacpac and AmberTools' sqm, deviating most in smaller molecules between 4 and 9 heavy atoms, in negatively charged molecules, and areas of chemistry underrepresented in the training set, such as particular sulfur- and phosphorus-containing functional groups. We further present the development and performance of Sage 2.3.0, which has both Lennard-Jones and valence parameters that are retrained to be consistent with neural network charges for the first time. Benchmarks spanning gas-phase geometry optimization through protein-ligand binding free energies show Sage 2.3.0 performs comparably to earlier Sage releases, with modest improvements in condensed-phase properties and a slight decrease in nonaqueous solvation free energy accuracy. As with other OpenFF force fields, Sage 2.3.0 was validated in protein-ligand benchmarks to be compatible with Amber protein force fields. All data are publicly available, along with scripts and environments for reproducing the training and benchmarking of AshGC and Sage 2.3.0 at https://github.com/openforcefield/ashgc-v1.0-fit and https://github.com/openforcefield/ash-sage-rc2 respectively.

Assembly-line enzymes carry out multi-step synthesis of diverse metabolites by using a handful of catalytic motifs in which minor structural differences control substrate specificity and reaction order. Here we examine differences in substrate binding to FabB and FabF, the two β-ketoacyl-ACP synthases (KSs) responsible for fatty acid elongation in Escherichia coli , by exploring a peculiar mutational effect. In FabB, a blocking mutation in the acyl binding pocket yields a shifted, but broad product profile, while in FabF, the same mutation disrupts the binding of acyl chains longer than eight carbons (C8). X-ray crystal structures of the FabB mutant provide an explanation: a second, previously unobserved binding pocket allows medium-to-long acyl chains (≥ C8) to bind with an alternate conformation. Molecular simulations suggest that this pocket is more stable in FabB than in FabF, where mutations reduce the catalytic competency of longer chains instead of shifting them to the alternate pocket. Our findings indicate that homologous KSs differ not only in their primary binding sites but also in the availability of alternative binding modes that can buffer against mutational effects and enable functional diversification.

Nitazenes are an emergent class of synthetic opioids that often rival or exceed fentanyl in their potency. These compounds have been detected internationally in illicit drugs and are the cause of increasing numbers of hospitalizations and overdoses. New analogs are consistently released, making detection challenging — new ways of testing a wide range of nitazenes and their metabolic products are urgently needed. Here, we develop a computational protocol to redesign the plant abscisic acid receptor PYR1 to bind diverse nitazenes and maintain its dynamic transduction mechanism. The best design has a low nanomolar limit of detection in vitro against nitazene and menitazene. Deep mutational scanning yielded sensors able to recognize a range of clinically relevant nitazenes and the common metabolic byproduct in a complex biological matrix with limited cross-specificity against unrelated opioids. Application of protein design tools on privileged receptors like PYR1 may yield general sensors for a wide range of applications in vitro and in vivo.

We report a new charge model and a new general small molecule force field. Here, we address the development and benchmarking of both the Open Force Field (OpenFF) AshGC charge model, as well as the Sage 2.3.0 small molecule force field for drug-like molecules. AshGC is a graph neural network-based method for efficient assignment of charges to both small and large molecules, designed to produce charges of semi- empirical quality at low cost. We describe the model, its architecture, featurization, training process, and performance. AshGC largely generates charges within the range of other accepted AM1-BCC backends such as OpenEye’s oequacpac and AmberTools’ sqm, but we note areas of poor performance such as in smaller molecules and in charged molecules, as well as areas of chemistry underrepresented in the training set. We fur- ther describe the development and performance of Sage 2.3.0, where Lennard-Jones and valence parameters are re-trained with neural network charges for the first time. Our benchmark suite of common force field tasks ranges from optimizing molecular structures in vacuum to protein-ligand binding free energies. Overall, Sage 2.3.0 performs comparably to previous Sage force fields, although we report minor improvements in the prediction of condensed phase properties and a minor decrease in performance in the prediction of non-aqueous solvation free energies. As with other OpenFF force fields, Sage 2.3.0 was validated in protein-ligand benchmarks to be compatible with Amber protein force fields. All data are publicly available, along with scripts, environments, and complete details for reproducing the training and benchmarking of AshGC and Sage 2.3.0 at https://github.com/openforcefield/ashgc-v1.0-fit and github.com/openforcefield/ash-sage-rc2 respectively.

Accurate free energy calculations are essential for predicting thermodynamic stability across complex molecular materials. Classical approaches such as the Einstein crystal method (ECM) provide benchmark accuracy but are computationally prohibitive for flexible, high-dimensional systems. Here we develop a scalable framework that couples targeted free energy perturbation with normalizing flow neural networks trained to learn bijective mappings between Boltzmann ensembles and analytical reference states. This enables robust lattice free energy estimates directly from molecular dynamics trajectories, eliminating the need for intermediate alchemical sampling. The method handles supercells with up to 3 x 10^3 degrees of freedom and supports multi-polymorph training within a single model. Validation across succinic acid, Veliparib, and Mivebresib demonstrates excellent agreement with ECM at significantly lower cost. By combining statistical mechanics and machine-learning expressivity, this approach advances free energy prediction toward automation and high-throughput applications in molecular and materials design.

In reverse osmosis (RO) and nanofiltration (NF) membranes, the polymer structure and interactions with solvent and solutes dictate the permeability and selectivity. However, these interactions have not been fully characterized within hydrated polymer membranes. In this study, we elucidate the local atomic neighborhood around ions within a RO membrane using molecular dynamics (MD). We built a MD model of a RO membrane closely following experimental synthesis and performed long time scale simulations of ions moving within the polymer. We find that the ion-oxygen nearest neighbor distance within the membrane is essentially the same as in solution, indicating that ions coordinate similarly in the confined membrane as in water. However, we do find that the average coordination number decreases in the polymer, which we attribute primarily to shifting the outer portion of the solvation shell beyond the cutoff, rather than being entirely stripped away. We find that cations bind tightly to both the carboxylate and amide oxygen atoms within the membrane. Even in ionized membranes, binding to amide oxygen atoms appears to play a substantial role in hindering ion mobility. Finally, we find that commonly used measures of ionic solvation structure such as coordination numbers do not fully capture the solvation structure, and we explore other measures such as the chemical composition of the nearest neighbors and the radial distribution function.