A string of surprising advances suggests usable quantum computers could be here in a decade.

We present a distilled multi-time-step (DMTS) strategy to accelerate molecular dynamics simulations using foundation neural network models. DMTS uses a dual-level neural network, where the target accurate potential is coupled to a simpler but faster model obtained via a distillation process. The 3.5 Å cutoff distilled model is sufficient to capture the fast-varying forces, i.e., mainly bonded interactions, from the accurate potential, allowing its use in a reversible reference system propagator algorithm (RESPA)-like formalism. The approach conserves accuracy, preserving both static and dynamic properties, while enabling us to evaluate the costly model only every 3 to 6 fs depending on the system. Consequently, large simulation speedups over standard 1 fs integration are observed: nearly 4-fold in homogeneous systems and 3-fold in large solvated proteins through leveraging active learning for enhanced stability. Such a strategy is applicable to any neural network potential and reduces the performance gap with classical force fields.

With so little money to go round, the costs of competing for grants can exceed what the grants are worth. When that happens, nobody wins.

Coarse-grained molecular dynamics simulations, such as those performed with the recently parametrized Martini 3 force field, simplify molecular models and enable the study of larger systems over longer time scales. With this new implementation, Martini 3 allows more bead types and sizes, becoming more amenable to studying dynamical phenomena involving small molecules such as protein–ligand interactions and membrane permeation. However, while solutions existed to automatically model small molecules using the previous iteration of the Martini force field, there is no simple way to generate such molecules for Martini 3 yet. Here, we introduce Auto-MartiniM3, an advanced and updated version of the Auto-Martini program designed to automate the coarse-graining of small molecules to be used with the Martini 3 force field. We validated our approach by modeling 81 simple molecules from the Martini Database and comparing their structural and thermodynamic properties with those obtained from models designed by Martini experts. Additionally, we assessed the behavior of Auto-MartiniM3-generated models by calculating solute translocation and free energy across lipid bilayers. We also evaluated more complex molecules such as caffeine by testing its binding to the adenosine A2A receptor. Finally, our results from deploying Auto-MartiniM3 on a large data set of molecular fragments demonstrate that this program can become a tool of choice for fast, high-throughput creation of coarse-grained models of small molecules, offering a good balance between automation and accuracy. Auto-MartiniM3 source code is freely available at https://github.com/Martini-Force-Field-Initiative/Automartini_M3.

Machine learning interatomic potentials (MLIPs) have become powerful tools to extend molecular simulations beyond the limits of quantum methods, offering near-quantum accuracy at much lower computatio...

FeNNol Pretrained Models Collection. Contribute to FeNNol-tools/FeNNol-PMC development by creating an account on GitHub.

L'an dernier, les présidents d'universités étaient déjà montés au créneau en décembre pour dénoncer les restrictions budgétaires demandées par le gouvernement, après déjà plusieurs années de sous-fina...

Researchers have developed a new mathematical strategy that efficiently identifies the essential building blocks of complex systems, dramatically improving computational power for applications ranging...