(16j) Unsupervised Optimization of Non-Bonded Parameters in Molecular Mechanics Force Fields Using the Particle Swarm Method.

In this work, we describe an automated process for the optimization of non-bonded interactions in molecular mechanics force fields to reproduce selected experimental data, such as vapor-liquid coexistence densities. The particle swam optimization (PSO) method[1] is combined with isobaric-isothermal ensemble Monte Carlo simulations at two-three state points to provide an estimate of the optimal parameters. The PSO method allows for the efficient evaluation of a wide range of parameter values, enhancing the probability of finding a global minimum. Additionally, with the PSO method, it is possible to perform multi-dimensional optimization of parameters, producing insights into relationships between various parameters and physical properties that may be missed using lower dimensional optimization strategies. These estimated parameters as used as input to Gibbs ensemble or grand canonical histogram reweighting Monte Carlo simulations. Parameters may be further optimized by reweighting in parameter space with the Multistate Bennett Acceptance Ratio (MBAR) method[2]. The methodology is applied to the optimization of non-bonded Mie potential parameters for three-site water models, ethanol and benzene. Integration of the PSO software and GOMC[3] is discussed.

References:

[1] Eberhart, RC,Shi, YH. Particle swarm optimization: Developments, applications and resources. Ieee C Evol Computat, 2001: 81-6.

[2] Shirts, MR,Chodera, JD. Statistically optimal analysis of samples from multiple equilibrium states. J. Chem. Phys., 2008; 129(12).

[3] Nejahi, Y, Barhaghi, MS, Mick, J, Jackman, B, Rushaidat, K, Li, YZ, et al. GOMC: GPU Optimized Monte Carlo for the simulation of phase equilibria and physical properties of complex fluids. Softwarex, 2019; 9: 20-7.