Empirical optimization of molecular simulation force fields by Bayesian inference

Abstract The demands on the accuracy of force fields for classical molecular dynamics simulations are steadily growing as larger and more complex systems studied over longer times. One way to meet these is hand learning their parameters machines in a systematic (semi)automatic manner. Doing so, we can take full advantage exascale computing, increasing availability experimental data, advances quantum mechanical computations calculation observables from ensembles. Here, discuss illustrate challenges one faces this endeavor explore forward by adapting Bayesian inference ensembles (BioEn) method [Hummer Köfinger, J. Chem. Phys. (2015)] field parameterization. In (BioFF) developed here, optimization problem regularized simplified prior an entropic acting ensemble. latter compensates unavoidable simplifications parameter prior. We determine optimal using iterative predictor–corrector approach, which run simulations, reference ensemble weighted histogram analysis (WHAM), update according BioFF posterior. approach simple polymer model, distance between two labeled sites observable. By systematically resolving issues, instead just reweighting structural ensemble, corrections extend not included reweighting. envision future formalized, systematic, machine-learning effort that incorporates wide range data experiment high-level chemical calculations, takes computing resources. Graphic abstract