Machine Learning Quantum Mechanical/Molecular Mechanical Potentials: Evaluating Transferability in Dihydrofolate Reductase-Catalyzed Reactions

Journal of Chemical Theory and Computation, Год журнала: 2025, Номер unknown

Опубликована: Янв. 15, 2025

Integrating machine learning potentials (MLPs) with quantum mechanical/molecular mechanical (QM/MM) free energy simulations has emerged as a powerful approach for studying enzymatic catalysis. However, its practical application been hindered by the time-consuming process of generating necessary training, validation, and test data MLP models through QM/MM simulations. Furthermore, entire needs to be repeated each specific enzyme system reaction. To overcome this bottleneck, it is required that trained MLPs exhibit transferability across different environments reacting species, thereby eliminating need retraining new variant. In study, we explore potential evaluating pretrained ΔMLP model mutations within MM environment using QM/MM-based ML architecture developed Pan, X. J. Chem. Theory Comput. 2021, 17(9), 5745–5758. The study includes scenarios such single point substitutions, homologous from even transition an aqueous environment, where last two systems have substantially used in training. results show effectively captures predicts effects on electrostatic interactions, producing reliable profiles enzyme-catalyzed reactions without retraining. also identified notable limitations transferability, particularly when transitioning water-rich environments. Overall, demonstrates robustness Pan et al.'s diverse systems, well further research development more sophisticated training methods.

Язык: Английский

---

DeePMD-GNN: A DeePMD-kit Plugin for External Graph Neural Network Potentials

Journal of Chemical Information and Modeling, Год журнала: 2025, Номер unknown

Опубликована: Март 27, 2025

Machine learning potentials (MLPs) have revolutionized molecular simulation by providing efficient and accurate models for predicting atomic interactions. MLPs continue to advance had profound impact in applications that include drug discovery, enzyme catalysis, materials design. The current landscape of MLP software presents challenges due the limited interoperability between packages, which can lead inconsistent benchmarking practices necessitates separate interfaces with dynamics (MD) software. To address these issues, we present DeePMD-GNN, a plugin DeePMD-kit framework extends its capabilities support external graph neural network (GNN) potentials.DeePMD-GNN enables seamless integration popular GNN-based models, such as NequIP MACE, within ecosystem. Furthermore, new infrastructure allows GNN be used combined quantum mechanical/molecular mechanical (QM/MM) using range corrected ΔMLP formalism.We demonstrate application DeePMD-GNN performing benchmark calculations NequIP, DPA-2 developed under consistent training conditions ensure fair comparison.

Язык: Английский

---

Neural Network Potential with Multiresolution Approach Enables Accurate Prediction of Reaction Free Energies in Solution

Journal of the American Chemical Society, Год журнала: 2025, Номер unknown

Опубликована: Фев. 17, 2025

We present the design and implementation of a novel neural network potential (NNP) its combination with an electrostatic embedding scheme, commonly used within context hybrid quantum-mechanical/molecular-mechanical (QM/MM) simulations. Substitution computationally expensive QM Hamiltonian by NNP same accuracy largely reduces computational cost enables efficient sampling in prospective MD simulations, main limitation faced traditional QM/MM setups. The model relies on recently introduced anisotropic message passing (AMP) formalism to compute atomic interactions encode symmetries found systems. AMP is shown be highly terms both data costs can readily scaled sample systems involving more than 350 solute 40,000 solvent atoms for hundreds nanoseconds using umbrella sampling. Most deviations predictions from underlying DFT ground truth lie chemical (4.184 kJ mol–1). performance broad applicability our approach are showcased calculating free-energy surface alanine dipeptide, preferred ligation states nickel phosphine complexes, dissociation free energies charged pyridine quinoline dimers. Results this ML/MM show excellent agreement experimental reach most cases. In contrast, calculated static calculations paired implicit models or simulations cheaper semiempirical methods up ten times higher deviation sometimes even fail reproduce qualitative trends.

Язык: Английский

---

DeePMD-kit v3: A Multiple-Backend Framework for Machine Learning Potentials

Journal of Chemical Theory and Computation, Год журнала: 2025, Номер unknown

Опубликована: Май 2, 2025

In recent years, machine learning potentials (MLPs) have become indispensable tools in physics, chemistry, and materials science, driving the development of software packages for molecular dynamics (MD) simulations related applications. These packages, typically built on specific frameworks, such as TensorFlow, PyTorch, or JAX, face integration challenges when advanced applications demand communication across different frameworks. The previous TensorFlow-based implementation DeePMD-kit exemplified these limitations. this work, we introduce version 3, a significant update featuring multibackend framework that supports PaddlePaddle backends, demonstrate versatility architecture through other MLP differentiable force fields. This allows seamless back-end switching with minimal modifications, enabling users developers to integrate using innovation facilitates more complex interoperable workflows, paving way broader MLPs scientific research.

Язык: Английский

---

Revolutionizing Molecular Design for Innovative Therapeutic Applications through Artificial Intelligence

Molecules, Год журнала: 2024, Номер 29(19), С. 4626 - 4626

Опубликована: Сен. 29, 2024

The field of computational protein engineering has been transformed by recent advancements in machine learning, artificial intelligence, and molecular modeling, enabling the design proteins with unprecedented precision functionality. Computational methods now play a crucial role enhancing stability, activity, specificity for diverse applications biotechnology medicine. Techniques such as deep reinforcement transfer learning have dramatically improved structure prediction, optimization binding affinities, enzyme design. These innovations streamlined process allowing rapid generation targeted libraries, reducing experimental sampling, rational tailored properties. Furthermore, integration approaches high-throughput techniques facilitated development multifunctional novel therapeutics. However, challenges remain bridging gap between predictions validation addressing ethical concerns related to AI-driven This review provides comprehensive overview current state future directions engineering, emphasizing their transformative potential creating next-generation biologics advancing synthetic biology.

Язык: Английский

---

PM6-ML: The Synergy of Semiempirical Quantum Chemistry and Machine Learning Transformed into a Practical Computational Method

Journal of Chemical Theory and Computation, Год журнала: 2025, Номер unknown

Опубликована: Янв. 3, 2025

Machine learning (ML) methods offer a promising route to the construction of universal molecular potentials with high accuracy and low computational cost. It is becoming evident that integrating physical principles into these models, or utilizing them in Δ-ML scheme, significantly enhances their robustness transferability. This paper introduces PM6-ML, method synergizes semiempirical quantum-mechanical (SQM) PM6 state-of-the-art ML potential applied as correction. The demonstrates superior performance over standalone SQM approaches covers broader chemical space than its predecessors. scalable systems thousands atoms, which makes it applicable large biomolecular systems. Extensive benchmarking confirms PM6-ML's robustness. Its practical application facilitated by direct interface MOPAC. code parameters are available at https://github.com/Honza-R/mopac-ml.

Язык: Английский

---

Advances in the Simulations of Enzyme Reactivity in the Dawn of the Artificial Intelligence Age

Wiley Interdisciplinary Reviews Computational Molecular Science, Год журнала: 2025, Номер 15(1)

Опубликована: Янв. 1, 2025

ABSTRACT The study of natural enzyme catalytic processes at a molecular level can provide essential information for rational design new enzymes, to be applied in more efficient and environmentally friendly industrial processes. use computational tools, combined with experimental techniques, is providing outstanding milestones the last decades. However, apart from complexity associated nature these large flexible biomolecular machines, full catalyzed process involves different physical chemical steps. Consequently, point view, deep understanding every single step requires selection proper technique get reliable, robust useful results. In this article, we summarize techniques their process, including conformational diversity, allostery those steps, as well enzymes. Because impact artificial intelligence all aspects science during years, special attention has been methods based on foundations some selected recent applications.

Язык: Английский

---

Modeling Enzyme Reaction and Mutation by Direct Machine Learning/Molecular Mechanics Simulations

Journal of Chemical Theory and Computation, Год журнала: 2025, Номер unknown

Опубликована: Апрель 24, 2025

Accurately modeling enzyme reactions through direct machine learning/molecular mechanics simulations remains challenging in describing the electrostatic coupling between QM and MM subsystems. In this work, we proposed a reweighting ME (mechanic embedding) REANN (recursively embedded atom neural network) method that trains potential point charges of subsystem vacuo. The charge equilibration approach has been encoded into to ensure conservation total subsystem. Electrostatic is measured by charges, polarization on can be corrected thermodynamic perturbation after molecular dynamics simulations. We first constructed surfaces energy for acylation cyclooxygenase-1 (COX-1) cyclooxygenase-2 (COX-2) aspirin. These allowed us reproduce free curves B3LYP/MM-MD with chemical accuracy. Subsequently, they were successfully applied R513A COX-2, reproducing barrier simulated B3LYP/MM MD difference less than 0.5 kcal mol-1 speedup 80-fold, revealing our predict activity mutants accurately rapidly. This expected virtual screening future.

Язык: Английский

---

The QDπ dataset, training data for drug-like molecules and biopolymer fragments and their interactions

Scientific Data, Год журнала: 2025, Номер 12(1)

Опубликована: Апрель 25, 2025

The development of universal machine learning potentials (MLP) for small organic and drug-like molecules requires large, accurate datasets that span diverse chemical spaces. In this study, we introduce the QDπ dataset which incorporates data taken from several datasets. We use a query-by-committee active strategy to extract large maximize diversity avoid redundancy as relevant neural network training construct dataset. only 1.6 million structures express 13 elements various source at ωB97M-D3(BJ)/def2-TZVPPD level theory. enables creation flexible target loss functions drug discovery, including information-dense sets relative conformational energies barriers, intermolecular interactions, tautomers protonation compounds biomolecular fragments. It is hope high information density contained in will provide valuable resource new MLPs discovery.

Язык: Английский

---

Multiobjective Evolutionary Strategy for Improving Semiempirical Hamiltonians in the Study of Enzymatic Reactions at the QM/MM Level of Theory

Journal of Chemical Theory and Computation, Год журнала: 2025, Номер unknown

Опубликована: Май 7, 2025

Quantum mechanics/molecular mechanics (QM/MM) simulations are crucial for understanding enzymatic reactions, but their accuracy depends heavily on the quantum-mechanical method used. Semiempirical methods offer computational efficiency often struggle with in complex systems. This work presents a novel multiobjective evolutionary strategy optimizing semiempirical Hamiltonians, specifically designed to enhance performance QM/MM while remaining broadly applicable condensed-phase Our methodology combines automated parameter optimization, targeting ab initio or density functional theory (DFT)-reference potential energy surfaces, atomic charges, and gradients, comprehensive validation through minimum free path (MFEP) calculations. To demonstrate its effectiveness, we applied our approach improve GFN2-xTB Hamiltonian using two systems that involve hydride transfer reactions where activation barrier is severely underestimated: Crotonyl-CoA carboxylase/reductase (CCR) dihydrofolate reductase (DHFR). The optimized parameters showed significant improvements reproducing closely matching higher-level DFT Through an efficient two-stage optimization process, first developed CCR reaction data, then refined these DHFR by incorporating targeted set of additional training geometries. strategic minimized cost achieving accurate descriptions both systems, as validated Adaptive String Method (ASM). represents study larger longer time scales, applications mechanism studies, drug design, enzyme engineering.

Язык: Английский

---