Impact of replicas and simulation length on in silico behaviors of a protein domain.

Molecular dynamics (MD) simulations are a powerful tool for life sciences, valuable for their ability to capture atomic-level behavior of molecules over time. To advance knowledge on reasonable timescales, researchers must optimize the amount of useful information extracted from simulation data while frugally managing computational resources. They must balance trajectory lengths and number of replicas, with the aim of maximizing conformational sampling. Identifying this balance is not always intuitive, and lack of standardization among researchers produces variability in results from MD measurements. We investigate how changes in simulation length and replica numbers impact this variability. Using a 231-residue domain, we compare measurements from independent trajectories to a benchmark trajectory of 3x1000-ns replicates. We simulate 27 protein-ligand complexes, allowing us to compare ligand-specific rankings of complexes across replicas. We reveal that some MD measurements are reliably ranked by single trajectories, while others are not. We uncover similar variability in the effects of trajectory lengths on measurements. Our findings suggest that a one-size-fits-all approach to MD simulations is not ideal, and depending on the intended measurements and research question, it is sometimes advantageous to prioritize longer trajectories over multiple replicas. This work provides important considerations for researchers while designing simulation studies.