Machine Learning Directed Optimization of Classical Molecular Modeling Force Fields

Befort, Bridgette; DeFever, Ryan S.; Tow, Garrett M.; Dowling, Alexander W.; Maginn, Edward J.

doi:10.1021/acs.jcim.1c00448

Cited by 47 publications

(87 citation statements)

References 84 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…One of the discussed limitations for these ab-initio-based models is that they remain computationally expensive compared to classical FFs and MD simulations. Another recent work by Befort et al describes an alternative top-down approach to ML-parametrized FFs . Their workflow incorporates experimental validation, domain knowledge, and surrogate models to perform an extensive optimization of Lennard-Jones parameters for two different small molecules with an objective to reproduce either the vapor–liquid equilibrium or crystal structure.…”

Section: Opportunities For High-throughput Force Field Developmentmentioning

confidence: 99%

Strategies for the Development of Conjugated Polymer Molecular Dynamics Force Fields Validated with Neutron and X-ray Scattering

et al. 2021

View full text Add to dashboard Cite

Conjugated polymers (CPs) enable a wide range of lightweight, lower cost, and flexible organic electronic devices, but a thorough understanding of relationships between molecular structure and dynamics and electronic performance is critical for improved device efficiencies and for new technologies. Molecular dynamics (MD) simulations offer in silico insight into this relationship, but their accuracy relies on the approach used to develop the model's parameters or force field (FF). In this Perspective, we first review current FFs for CPs and find that most of the models implement an arduous reparameterization of inter-ring torsion potentials and partial charges of classical FFs. However, there are few FFs outside of simple CP molecules, e.g., polythiophenes, that have been developed over the last two decades. There is also limited reparameterization of other parameters, such as nonbonded Lennard-Jones interactions, which we find to be directly influenced by conjugation in these materials. We further provide a discussion on experimental validation of MD FFs, with emphasis on neutron and X-ray scattering. We define multiple ways in which various scattering methods can be directly compared to results of MD simulations, providing a powerful experimental validation metric of local structure and dynamics at relevant length and time scales to charge transport mechanisms in CPs. Finally, we offer a perspective on the use of neutron scattering with machine learning to enable high-throughput parametrization of accurate and experimentally validated CP FFs enabled not only by the ongoing advancements in computational chemistry, data science, and high-performance computing but also using oligomers as proxies for longer polymer chains during FF development.

show abstract

Section: Opportunities For High-throughput Force Field Developmentmentioning

confidence: 99%

Strategies for the Development of Conjugated Polymer Molecular Dynamics Force Fields Validated with Neutron and X-ray Scattering

et al. 2021

View full text Add to dashboard Cite

show abstract

“…While these surrogates are not perfect imitations of the high-level response, we can construct them to a reasonable level of accuracy with a limited number of expensive evaluations. In the context of molecular simulation parameter optimization, Befort et al [45] demonstrated a method of optimizing LJ parameters by building GP surrogates based on physical properties and applied this method to several hydrofluorocarbons as well as ammonium perchlorate.…”

Section: Surrogate Modeling Can Accelerate Non-bonded Trainingmentioning

confidence: 99%

Using physical property surrogate models to perform multi-fidelity global optimization of force field parameters

Madin

Shirts

2022

Preprint

View full text Add to dashboard Cite

Dispersion-repulsion interactions, commonly represented in atomistic force fields by the Lennard-Jones (LJ) potential, play an important role in the accuracy of molecular simulations. Training the force field parameters used in the LJ potential is challenging, generally requiring adjustment based on simulations of macroscopic physical properties. The computational expense of these simulations limits the size of training data set and number of optimization steps that can be taken, often requiring modelers to perform optimizations within a local parameter region. To allow for global LJ parameter optimization against large training sets, we introduce a multi-fidelity optimization technique which uses Gaussian process surrogate modeling to build inexpensive models of physical properties as a function of LJ parameters. This allows for fast evaluation of objective functions, greatly accelerating searches over parameter space. We use an iterative framework which performs global optimization at the surrogate level, followed by validation at the simulation level and surrogate refinement. Using this technique on two previously studied training sets, containing up to 195 physical property targets, we refit a subset of the LJ parameters for the OpenFF 1.0.0 ``Parsley'' force field. We demonstrate that this multi-fidelity technique can find improved parameter sets compared to a purely simulation-based optimization by searching more broadly and escaping local minima. In most cases, these parameter sets are transferable to other similar molecules in a test set. This multi-fidelity technique provides a platform for fast optimization against physical properties that can be refined and applied in multiple ways to the development of molecular models.

show abstract

“…To apply this technique to an arbitrary force field, one will usually need to construct such surrogate models for different properties. Common techniques to build these surrogate models might include reweighting [46], Gaussian processes [47], and machine learning methods [36]. Since the methods needed to construct such a model depend substantially on the property and the parameters of interest, that question is beyond the scope of this study; we focus only on applying Bayesian inference given a surrogate model.…”

Section: Surrogate Models For 2cljq Outputmentioning

confidence: 99%

“…This is a challenging task, but should be possible by simulating properties of interest for multiple parameter sets, and then using modeling techniques well suited to sparse data, such as Gaussian processes (GP) [73,74] regression. Befort et al [47] recently had success using GPs to build physical property surrogate models based on small molecule force fields. These methods could be enhanced with reweighting techniques like MBAR on simulated points to add gradient information to the surrogate models.…”

Section: Challenges Of Implementationmentioning

confidence: 99%

Bayesian inference-driven model parameterization and model selection for 2CLJQ fluid models

Madin¹,

Boothroyd²,

Messerly³

et al. 2021

Preprint

View full text Add to dashboard Cite

A high level of physical detail in a molecular model improves its ability to perform high accuracy simulations, but can also significantly affect its complexity and computational cost. In some situations, it is worthwhile to add additional complexity to a model to capture properties of interest; in others, additional complexity is unnecessary and can make simulations computationally infeasible. In this work we demonstrate the use of Bayes factors for molecular model selection, using Monte Carlo sampling techniques to evaluate the evidence for different levels of complexity in the two-centered Lennard-Jones + quadrupole (2CLJQ) fluid model. Examining three levels of nested model complexity, we demonstrate that the use of variable quadrupole and bond length parameters in this model framework is justified only sometimes. We also explore the effect of the Bayesian prior distribution on the Bayes factors, as well as ways to propose meaningful prior distributions. This Bayesian Markov Chain Monte Carlo (MCMC) process is enabled by the use of analytical surrogate models that accurately approximate the physical properties of interest. This work paves the way for further atomistic model selection work via Bayesian inference and surrogate modeling.

show abstract

Machine Learning Directed Optimization of Classical Molecular Modeling Force Fields

Cited by 47 publications

References 84 publications

Strategies for the Development of Conjugated Polymer Molecular Dynamics Force Fields Validated with Neutron and X-ray Scattering

Strategies for the Development of Conjugated Polymer Molecular Dynamics Force Fields Validated with Neutron and X-ray Scattering

Using physical property surrogate models to perform multi-fidelity global optimization of force field parameters

Bayesian inference-driven model parameterization and model selection for 2CLJQ fluid models

Contact Info

Product

Resources

About