Scribd
Upload
6 views1 page

Abstract EDB Ai24

The document discusses advancements in molecular dynamics simulations, highlighting the shift from traditional empirical force fields to machine learning approaches. It introduces the kernel-based minimal distributed charge model (kMDCM), which enhances electrostatic modeling by incorporating anisotropic and conformationally dependent features. The kMDCM is integrated with both empirical and machine learning force fields, improving accuracy and efficiency in molecular simulations, and is now available in the CHARMM package.

Uploaded by

4boitter
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
6 views1 page

Abstract EDB Ai24

The document discusses advancements in molecular dynamics simulations, highlighting the shift from traditional empirical force fields to machine learning approaches. It introduces the kernel-based minimal distributed charge model (kMDCM), which enhances electrostatic modeling by incorporating anisotropic and conformationally dependent features. The kMDCM is integrated with both empirical and machine learning force fields, improving accuracy and efficiency in molecular simulations, and is now available in the CHARMM package.

Uploaded by

4boitter
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 1

SCS Symposium on AI 2024 November 27, 2024, University of Fribourg

Bridging Empirical and Machine Learning Force Fields with Anisotropic,


Conformationally Dependent Electrostatic Models
Eric D. Boittier, Markus Meuwly

Department of Chemistry, University of Basel, Klingelbergstrasse 80, CH-4056 Basel, Switzerland


ericdavid.boittier@unibas.ch

The last decade has seen a transformative shift in the landscape of molecular dynamics (MD)
simulations. Traditional empirical force fields (FFs), with carefully crafted functional forms and
parameter sets, contrast sharply with today’s machine learning (ML) approaches, which use data-driven
interpolation to approximate the potential energy surface with improved accuracy. While ML force fields
(MLFFs) can capture complex interactions beyond the reach of empirical models, they often lack the
physical priors necessary for reliable extrapolation in data-scarce regions. Another significant challenge
is the treatment of long-range interactions, as message-passing neural networks (MPNNs) rely on
aggregating local atomic environments and can lose long-range information. Conversely,
computationally efficient empirical FFs use fixed partial charges, which fail to capture conformational
sensitivity and electrostatic anisotropy.

To address these limitations, we present the kernel-based minimal distributed charge model (kMDCM),
a conformationally dependent electrostatic potential (ESP) model optimized for molecular simulations.
kMDCM introduces anisotropy through adaptive off-center charges, accurately modeling intramolecular
charge redistribution based on molecular geometry at a level comparable to atomic multipoles. We
demonstrate its application in combination with empirical FFs and MLFFs. Additionally, we enhance the
parameterization of distributed charge models by integrating SE(3)-equivariant MPNNs, enabling better
generalization across molecular conformations and chemical space. Through a student-teacher
framework, we transfer the fidelity of equivariant neural networks to efficient kernel-based
interpolation, balancing speed and accuracy. These advancements are now accessible in the CHARMM
molecular dynamics package, offering a robust, physically informed approach for next-generation MD
simulations.

[1] Behler, J. Four Generations of High-Dimensional Neural Network Potentials. Chem. Rev. 2021, 121
(16), 10037–10072.
[2] Boittier, E.; Töpfer, K.; Devereux, M.; Meuwly, M. Kernel-Based Minimal Distributed Charges: A
Conformationally Dependent ESP-Model for Molecular Simulations. J. Chem. Theory Comput. 2024, 20
(18), 8088–8099.
[3] Töpfer, K.; Boittier, E.; Devereux, M.; Pasti, A.; Hamm, P.; Meuwly, M. Force Fields for Deep Eutectic
Mixtures: Application to Structure, Thermodynamics and 2D-Infrared Spectroscopy. J. Phys. Chem. B
2024.

You might also like