Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Grappa -- A Machine Learned Molecular Mechanics Force Field (2404.00050v2)

Published 25 Mar 2024 in physics.chem-ph, cs.LG, and physics.comp-ph

Abstract: Simulating large molecular systems over long timescales requires force fields that are both accurate and efficient. In recent years, E(3) equivariant neural networks have lifted the tension between computational efficiency and accuracy of force fields, but they are still several orders of magnitude more expensive than established molecular mechanics (MM) force fields. Here, we propose Grappa, a machine learning framework to predict MM parameters from the molecular graph, employing a graph attentional neural network and a transformer with symmetry-preserving positional encoding. The resulting Grappa force field outperformstabulated and machine-learned MM force fields in terms of accuracy at the same computational efficiency and can be used in existing Molecular Dynamics (MD) engines like GROMACS and OpenMM. It predicts energies and forces of small molecules, peptides, RNA and - showcasing its extensibility to uncharted regions of chemical space - radicals at state-of-the-art MM accuracy. We demonstrate Grappa's transferability to macromolecules in MD simulations from a small fast folding protein up to a whole virus particle. Our force field sets the stage for biomolecular simulations closer to chemical accuracy, but with the same computational cost as established protein force fields.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (38)
  1. Learning local equivariant representations for large-scale atomistic dynamics, 2022. URL https://arxiv.org/abs/2204.05249.
  2. Mace: Higher order equivariant message passing neural networks for fast and accurate force fields, 2023.
  3. Tensornet: Cartesian tensor representations for efficient learning of molecular potentials, 2023.
  4. Accurate machine learned quantum-mechanical force fields for biomolecular simulations, 2022.
  5. Equivariant message passing for the prediction of tensorial properties and molecular spectra, 2021. URL https://arxiv.org/abs/2102.03150.
  6. End-to-end differentiable construction of molecular mechanics force fields. Chemical Science, 13(41):12016–12033, 2022. doi: 10.1039/d2sc02739a. URL https://doi.org/10.1039%2Fd2sc02739a.
  7. Attention is all you need, 2023. URL https://arxiv.org/abs/1706.03762.
  8. Gromacs: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. SoftwareX, 1-2:19–25, 2015. ISSN 2352-7110.
  9. Openmm 4: A reusable, extensible, hardware independent library for high performance molecular simulation. Journal of Chemical Theory and Computation, 9(1):461–469, 2013. doi: 10.1021/ct300857j. URL https://doi.org/10.1021/ct300857j. PMID: 23316124.
  10. Espaloma-0.3.0: Machine-learned molecular mechanics force field for the simulation of protein-ligand systems and beyond, 2023.
  11. Graph attention networks, 2018. URL https://arxiv.org/abs/1710.10903.
  12. Deep residual learning for image recognition, 2015.
  13. Layer normalization, 2016.
  14. On the expressivity role of layernorm in transformers’ attention. arXiv preprint arXiv:2305.02582, 2023.
  15. Graph transformer networks, 2020.
  16. Fast, efficient generation of high-quality atomic charges. am1-bcc model: Ii. parameterization and validation. Journal of Computational Chemistry, 23(16):1623–1641, 2002. doi: https://doi.org/10.1002/jcc.10128. URL https://onlinelibrary.wiley.com/doi/abs/10.1002/jcc.10128.
  17. Improved side-chain torsion potentials for the amber ff99sb protein force field. Proteins, 78(8):1950–1958, Jun 2010. doi: 10.1002/prot.22711. URL https://doi.org/10.1002/prot.22711.
  18. Schnet: A continuous-filter convolutional neural network for modeling quantum interactions, 2017. URL https://arxiv.org/abs/1706.08566.
  19. Sobolev training for neural networks, 2017.
  20. Development and benchmarking of open force field 2.0.0 — the sage small molecule force field. ChemRxiv, 2022. doi: 10.1021/acs.jctc.3c00039. URL https://doi.org/10.1021/acs.jctc.3c00039. This content is a preprint and has not been peer-reviewed.
  21. A fast and high-quality charge model for the next generation general amber force field. The Journal of chemical physics, 153:114502, 09 2020. doi: 10.1063/5.0019056. URL https://doi.org/10.1063/5.0019056.
  22. ff14sb: Improving the accuracy of protein side chain and backbone parameters from ff99sb. Journal of Chemical Theory and Computation, 11(8):3696–3713, 2015. ISSN 1549-9618. doi: 10.1021/acs.jctc.5b00255. URL https://doi.org/10.1021/acs.jctc.5b00255.
  23. Refinement of the cornell et al. nucleic acids force field based on reference quantum chemical calculations of glycosidic torsion profiles. Journal of Chemical Theory and Computation, 7(9):2886–2902, 2011. ISSN 1549-9618. doi: 10.1021/ct200162x. URL https://doi.org/10.1021/ct200162x.
  24. Spice, a dataset of drug-like molecules and peptides for training machine learning potentials. Scientific Data, 10(11), 2023. URL https://doi.org/10.1038/s41597-022-01882-6.
  25. How fast-folding proteins fold. Science, 334(6055):517–520, 2011. doi: 10.1126/science.1208351. URL https://www.science.org/doi/abs/10.1126/science.1208351.
  26. Hybrid kinetic monte carlo/molecular dynamics simulations of bond scissions in proteins. Journal of Chemical Theory and Computation, 16(1):553–563, 2020. doi: 10.1021/acs.jctc.9b00786. URL https://doi.org/10.1021/acs.jctc.9b00786. PMID: 31738552.
  27. Pytorch: An imperative style, high-performance deep learning library, 2019.
  28. Deep graph library: A graph-centric, highly-performant package for graph neural networks, 2020.
  29. Fast and accurate deep network learning by exponential linear units (elus), 2016.
  30. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15:1929–1958, 06 2014.
  31. Adam: A method for stochastic optimization, 2017.
  32. Sgdr: Stochastic gradient descent with warm restarts, 2017.
  33. New formulation for derivatives of torsion angles and improper torsion angles in molecular mechanics: Elimination of singularities. Journal of Computational Chemistry, 17(9):1132–1141, 1996.
  34. Psi4 1.4: Open-source software for high-throughput quantum chemistry. J. Chem. Phys., 2020. doi: 10.1063/5.0006002.
  35. Berk Hess. P-LINCS: A parallel linear constraint solver for molecular simulation. 4(1):116–122. ISSN 1549-9618, 1549-9626. doi: 10.1021/ct700200b. URL https://pubs.acs.org/doi/10.1021/ct700200b.
  36. Canonical sampling through velocity rescaling. 126(1):014101. ISSN 0021-9606, 1089-7690. doi: 10.1063/1.2408420.
  37. M. Parrinello and A. Rahman. Polymorphic transitions in single crystals: A new molecular dynamics method. 52(12):7182–7190. ISSN 0021-8979. doi: 10.1063/1.328693. URL https://doi.org/10.1063/1.328693.
  38. Crystal structure of a ten-amino acid protein. Journal of the American Chemical Society, 130(46):15327–15331, 2008. doi: 10.1021/ja8030533. URL https://doi.org/10.1021/ja8030533. PMID: 18950166.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Leif Seute (3 papers)
  2. Eric Hartmann (1 paper)
  3. Jan Stühmer (13 papers)
  4. Frauke Gräter (4 papers)
Citations (2)
View on Semantic Scholar

Summary

Overview of Grappa: A Machine Learned Molecular Mechanics Force Field

The paper "Grappa - A Machine Learned Molecular Mechanics Force Field," presents an innovative approach to molecular mechanics (MM) by employing ML techniques to achieve state-of-the-art accuracy in simulating molecular systems. Traditionally, MM force fields have relied on physics-inspired functional forms, requiring significant simplification and limiting the accuracy of the simulations. This work stands out by introducing Grappa, a machine learning framework that embodies MM's inherent efficiency while significantly enhancing accuracy.

Grappa distinguishes itself by leveraging a graph attentional neural network and transformer architecture with symmetry-preserving positional encoding, eliminating the need for hand-crafted chemical features. This facilitates the prediction of MM parameters directly from the molecular graph, allowing it to traverse extensive chemical space with greater generalization capabilities. The architecture respects the permutation symmetries inherent in MM, further aligning with the physical realities of molecular interactions.

Methodology and Technical Contribution

Grappa uses graph attentional neural networks to predict atom embeddings, which represent local chemical environments. These embeddings are passed through symmetric transformers, mapping them to MM parameters with desired permutation symmetries. The architecture is designed to respect specific symmetries for different interaction types within a molecular graph, enhancing expressivity while ensuring physical significance.

Grappa demonstrates substantial improvements in terms of RMSE for energy and force predictions, outperforming both traditional MM force fields and existing machine-learned MM force fields such as Espaloma across varied datasets. This includes extensive datasets covering small molecules, peptides, and RNA. These results underline Grappa's superior accuracy and computational efficiency, akin to established MM force fields such as GROMACS and OpenMM.

Implications and Future Prospects

The advancements presented in Grappa have significant implications for the simulation of biomolecular systems, particularly in achieving closeness to quantum mechanical accuracy without the prohibitive computational costs typically associated with quantum chemical methods. Grappa's capabilities extend to protein simulations, showing exemplary performance in maintaining protein stability over nanosecond timescales and accurately recovering folded structures in MD simulations starting from unfolded states.

The elimination of hand-crafted features enables Grappa to generalize across uncharted chemical spaces, paving the way for extensions beyond conventional biomolecular simulations to areas like drug discovery and complex biochemical systems. The framework provides a robust platform for continued enhancement and integration of non-bonded parameters, which could further improve its utility across diverse molecular dynamics applications.

Conclusion

Grappa represents a significant step forward in the field of molecular simulation, seamlessly integrating machine learning with traditional MM frameworks to deliver high accuracy at feasible computational costs. The application of symmetric transformers and graph neural networks positions Grappa favorably for future advancements in both the theoretical understanding and practical applications of molecular mechanics force fields. Its success underscores the potential of ML in addressing longstanding challenges in computational chemistry and molecular dynamics.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown