Which corners to cut? Guidelines on choosing optimal settings to maximise sampling with limited computational resources

Despite the increasingly wide availability of computational resources, it is still challenging for researchers to perform comprehensive molecular dynamics (MD) simulations on an industrial scale. With ensemble approaches, great accuracy and precision are achievable at the cost of considerable computational efforts. There is a trade-off between accuracy, precision and the compute cost of performing ensembles. A question frequently raised is what is the most optimal way to perform MD based calculation of one or more properties of interest? Here, we use our findings from extensive ensemble molecular dynamics simulations of ligand-protein systems to underpin the recommendations we are making in cases where computational cost is an important consideration. We recommend performing a minimum of 5 replicas for the ESMACS-style protocol and 3 replicas (per λ window) for the TIES-like protocol, and the use of a stepwise procedure to reduce costs so as to minimise the loss of accuracy and precision of the results obtained.