Grappa – a machine learned molecular mechanics force field

Chemical Science, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 1, 2025

We propose Grappa, a machine learned molecular mechanics force field for proteins. operating on the graph, accurately predicts energies and forces agrees with experimental data such as J -couplings folding free energies.

Language: Английский

---

Scaling Graph Neural Networks to Large Proteins

Journal of Chemical Theory and Computation, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 6, 2025

Graph neural network (GNN) architectures have emerged as promising force field models, exhibiting high accuracy in predicting complex energies and forces based on atomic identities Cartesian coordinates. To expand the applicability of GNNs, machine learning fields more broadly, optimizing their computational efficiency is critical, especially for large biomolecular systems classical molecular dynamics simulations. In this study, we address key challenges existing GNN benchmarks by introducing a dataset, DISPEF, which comprises large, biologically relevant proteins. DISPEF includes 207,454 proteins with sizes up to 12,499 atoms features diverse chemical environments, spanning folded disordered regions. The implicit solvation free energies, used training targets, represent particularly challenging case due many-body nature, providing stringent test evaluating expressiveness models. We benchmark performance seven GNNs emphasizing importance directly accounting long-range interactions enhance model transferability. Additionally, present novel multiscale architecture, termed Schake, delivers transferable computationally efficient energy predictions Our findings offer valuable insights tools advancing protein modeling applications.

Language: Английский

---

The Open Force Field Initiative: Open Software and Open Science for Molecular Modeling

The Journal of Physical Chemistry B, Journal Year: 2024, Volume and Issue: unknown

Published: July 11, 2024

Force fields are a key component of physics-based molecular modeling, describing the energies and forces in system as function positions atoms molecules involved. Here, we provide review scientific status report on work Open Field (OpenFF) Initiative, which focuses science, infrastructure data required to build next generation biomolecular force fields. We introduce OpenFF Initiative related Consortium, describe its approach field development software, discuss accomplishments date well future plans. releases both software under open permissive licensing agreements enable rapid application, validation, extension, modification tools. lessons learned this new development. also highlight ways that other researchers can get involved, some recent successes outside taking advantage tools data.

Language: Английский

---

QupKake: Integrating Machine Learning and Quantum Chemistry for Micro-pK_a Predictions

Journal of Chemical Theory and Computation, Journal Year: 2024, Volume and Issue: 20(15), P. 6946 - 6956

Published: June 4, 2024

Accurate prediction of micro-pKa values is crucial for understanding and modulating the acidity basicity organic molecules, with applications in drug discovery, materials science, environmental chemistry. This work introduces QupKake, a novel method that combines graph neural network models semiempirical quantum mechanical (QM) features to achieve exceptional accuracy generalization prediction. QupKake outperforms state-of-the-art on variety benchmark data sets, root-mean-square errors between 0.5 0.8 pKa units five external test sets. Feature importance analysis reveals role QM both reaction site enumeration models. represents significant advancement prediction, offering powerful tool various chemistry beyond.

Language: Английский

---

Fine-tuning molecular mechanics force fields to experimental free energy measurements

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 8, 2025

Alchemical free energy methods using molecular mechanics (MM) force fields are essential tools for predicting thermodynamic properties of small molecules, especially via calculations that can estimate quantities relevant drug discovery such as affinities, selectivities, the impact target mutations, and ADMET properties. While traditional MM forcefields rely on hand-crafted, discrete atom types parameters, modern approaches based graph neural networks (GNNs) learn continuous embedding vectors represent chemical environments from which parameters be generated. Excitingly, GNN parameterization provide a fully end-to-end differentiable model offers possibility systematically improving these models experimental data. In this study, we treat pretrained field-here, espaloma-0.3.2-as foundation simulation fine-tune its charge limited hydration data, with goal assessing degree to improve prediction other related energies. We demonstrate highly efficient "one-shot fine-tuning" method an exponential (Zwanzig) reweighting estimator accuracy without need resimulate configurations. To achieve "one-shot" improvement, importance effective sample size (ESS) regularization strategies retain good overlap between initial fine-tuned fields. Moreover, show leveraging low-rank projections comparable improvements higher-dimensional in variety data-size regimes. Our results linearly-perturbative fine-tuning electrostatic data cost-effective strategy achieves state-of-the-art performance energies FreeSolv dataset.

Language: Английский

---

Molecular dynamics and machine learning unlock possibilities in beauty design—A perspective

AIP Advances, Journal Year: 2025, Volume and Issue: 15(1)

Published: Jan. 1, 2025

Computational molecular design—the endeavor to design molecules, with various missions, aided by machine learning and dynamics approaches—has been widely applied create valuable new entities, from small molecule therapeutics protein biologics. In the data regime, physics-based approaches model interaction between being designed proteins of key physiological functions, providing structural insights into mechanism. When abundant have collected, a quantitative structure–activity relationship can be more directly constructed experimental data, which distill guide next round experiment design. Machine methodologies also facilitate physical modeling, improving accuracy force fields extending them unseen chemical spaces enhancing sampling on conformational spaces. We argue that these techniques are mature enough not just extend longevity life but beauty it manifests. this Perspective, we review current frontiers in research development skincare products, as well statistical toolbox applicable addressing challenges industry. Feasible interdisciplinary projects proposed harness power tools innovative, effective, inexpensive products.

Language: Английский

---

ABCG2: A Milestone Charge Model for Accurate Solvation Free Energy Calculation

Journal of Chemical Theory and Computation, Journal Year: 2025, Volume and Issue: unknown

Published: March 11, 2025

In this report, we describe the development and validation of ABCG2, a new charge model with milestone free energy accuracy, while allowing instantaneous atomic assignment for arbitrary organic molecules. combination second-generation general AMBER force field (GAFF2), ABCG2 led to root-mean-square error (RMSE) 0.99 kcal/mol on hydration calculation all 642 solutes in FreeSolv database, first time meeting chemical accuracy threshold through physics-based molecular simulation against golden-standard data set. Against Minnesota Solvation Database, solvation 2068 pairs range diverse solvents an RMSE 0.89 kcal/mol. The 1913 points transfer energies from aqueous solution obtained 0.85 kcal/mol, corresponding 0.63 log units logP. benchmark densities neat liquids 1839 molecules heat vaporizations 874 achieved comparable performance default restrained electrostatic potential (RESP) method GAFF2. fluctuations assigned partial charges over different input conformations are demonstrated be much smaller than those RESP statistics 96 real drug results not only but also transferability generality GAFF2/ABCG2 combination.

Language: Английский

---

Polarizable Water Model with Ab Initio Neural Network Dynamic Charges and Spontaneous Charge Transfer

Journal of Chemical Theory and Computation, Journal Year: 2025, Volume and Issue: unknown

Published: March 29, 2025

Simulating water accurately has been a challenge due to the complexity of describing polarization and intermolecular charge transfer. Quantum mechanical (QM) electronic structures provide an accurate description in response local environments, which is nevertheless too expensive for large systems. In this study, we have developed polarizable model integrating Charge Model 5 atomic charges at level second-order Mo̷ller–Plesset perturbation theory, predicted by transferable neural network (ChargeNN) model. The spontaneous transfer explicitly accounted for, enabling precise treatment hydrogen bonds out-of-plane polarization. Our ChargeNN successfully reproduces various properties gas, liquid, solid phases. For example, correctly captures hydrogen-bond stretching peak bending-libration combination band, are absent spectra using fixed charges, highlighting significance Finally, molecular dynamical simulations liquid droplet with ∼4.5 nm radius reveal that strong interfacial electric fields concurrently induced partial collapse surface-to-interior study paves way QM-polarizable force fields, aiming large-scale high accuracy.

Language: Английский

---

On the design space between molecular mechanics and machine learning force fields

Applied Physics Reviews, Journal Year: 2025, Volume and Issue: 12(2)

Published: April 2, 2025

A force field as accurate quantum mechanics (QMs) and fast molecular (MMs), with which one can simulate a biomolecular system efficiently enough meaningfully to get quantitative insights, is among the most ardent dreams of biophysicists—a dream, nevertheless, not be fulfilled any time soon. Machine learning fields (MLFFs) represent meaningful endeavor in this direction, where differentiable neural functions are parametrized fit ab initio energies forces through automatic differentiation. We argue that, now, utility MLFF models no longer bottlenecked by accuracy but primarily their speed, well stability generalizability—many recent variants, on limited chemical spaces, have long surpassed 1 kcal/mol—the empirical threshold beyond realistic predictions possible—though still magnitudes slower than MM. Hoping kindle exploration design faster, albeit perhaps slightly less MLFFs, review, we focus our attention technical space (the speed-accuracy trade-off) between MM ML fields. After brief review building blocks (from machine learning-centric point view) either kind, discuss desired properties challenges now faced development community, survey efforts make more envision what next generation might look like.

Language: Английский

---

DASH: Dynamic Attention-Based Substructure Hierarchy for Partial Charge Assignment

Journal of Chemical Information and Modeling, Journal Year: 2023, Volume and Issue: 63(19), P. 6014 - 6028

Published: Sept. 22, 2023

We present a robust and computationally efficient approach for assigning partial charges of atoms in molecules. The method is based on hierarchical tree constructed from attention values extracted graph neural network (GNN), which was trained to predict atomic accurate quantum-mechanical (QM) calculations. resulting dynamic attention-based substructure hierarchy (DASH) provides fast assignment with the same accuracy as GNN itself, software-independent, can easily be integrated existing parametrization pipelines, shown Open force field (OpenFF). implementation DASH workflow, final tree, training set are available open source/open data public repositories.

Language: Английский

---