Force Field Parametrization of Metal Ions from Statistical Learning Techniques

Abstract

A novel statistical procedure has been developed to optimize the parameters of nonbonded force fields of metal ions in soft matter. The criterion for the optimization is the minimization of the deviations from ab initio forces and energies calculated for model systems. The method exploits the combination of the linear ridge regression and the cross-validation techniques with the differential evolution algorithm. Wide freedom in the choice of the functional form of the force fields is allowed since both linear and nonlinear parameters can be optimized. In order to maximize the information content of the data employed in the fitting procedure, the composition of the training set is entrusted to a combinatorial optimization algorithm which maximizes the dissimilarity of the included instances. The methodology has been validated using the force field parametrization of five metal ions (Zn2+, Ni2+, Mg2+, Ca2+, and Na+) in water as test cases.

The TensorMol-0.1 model chemistry: a neural network augmented with long-range physics

“…Statistical models from machine learning experienced growing popularity in many areas of chemistry, such as in reducing the cost of simulating chemical systems, 1 – 13 improving the accuracy of quantum methods, 14 – 22 generating force field parameters, 23 , 24 predicting molecular properties 25 – 32 and designing new materials. 33 – 38 Neural network model chemistries (NNMCs) are one of the most powerful methods among this class of models.…”

Tailor‐made computational protocols for precise characterization of small biological building blocks using QM and MM approaches

“…One such tool is Force Balance by Van Voorhis and coworkers; which in principle is able to simultaneously optimize all the components of a FF and provides a scheme to avoid over fitting based on a regularization procedure that cycles through several steps of sampling/reference data generation (e.g., QM calculations of forces and or energies)/optimization. Recently, Barone and coworkers proposed a method called Linear Ridge Regression—Differential Evolution (LRR‐DE) . LRR‐DE that uses a Tichonov Regularization scheme and cross validation to prevent over fitting of FF parameters and yield robust FF parameters.…”

Development of Nonbonded Models for Metal Cations Using the Electronic Continuum Correction

“…[9] Protocols aimed at optimizing such nonbonded models are based on the in silico reproduction of experimental properties, such as hydration free energy (ΔG, hereafter), ion-water oxygen distance (IOD) values and crystal lattice energies, [10][11][12][13][14] or using high-level quantum mechanics data as target values during the fitting procedure. [15][16][17] However, the parametrization procedure of ions concerns the Lennard-Jones parameters only; conversely, the atomic charge is set equal to the corresponding formal one, thus considering the ion itself in a vacuum. From a methodological point of view, this operation is in contrast with commonly used parametrization protocols for neutral molecules.…”

“…The easier strategy to include metal ions in classical MD simulations is represented by the nonbonded model, which treats the charged monoatomic species as a sphere interacting with the surrounding environment by means of Coulomb and Lennard‐Jones potentials . Protocols aimed at optimizing such nonbonded models are based on the in silico reproduction of experimental properties, such as hydration free energy (Δ G , hereafter), ion‐water oxygen distance (IOD) values and crystal lattice energies, or using high‐level quantum mechanics data as target values during the fitting procedure . However, the parametrization procedure of ions concerns the Lennard‐Jones parameters only; conversely, the atomic charge is set equal to the corresponding formal one, thus considering the ion itself in a vacuum.…”