When I go with Gromacs protein simulations, should I consider about performing long run (at least 100ns) or multiple runs (at least 2 runs)? Which one of these could be more reliable and why?
Usually, it is recommended (for different practical reasons, such as, avoiding having single very big file, limited available CPU time for very long time, storage place available in one location, etc.) to create multiple runs, where each run follows the previous run by creating 'restart files', which allow every run to use the initial state (coordinates and velocities, of both system and thermostat/barostat, if the case) from the previous run. This way is exactly the same as you would perform a single run! (Both, namely multiple runs with restart file and a single run, are exactly equivalent.)
There also exist other ways of running multiple short simulations (but, for completely different practical reasons); in that case the initial time configurations (which are different from each other) are generated from canonical distribution at the same temperature that you have fixed the simulation run. Using then particular methods for combining the results of individual simulations in order not to lose the sense of the total simulation run time.
A single plain (that is, not using any sampling enhancement) MD is the only way to understand innate intricacies of a physical process, such as, protein folding and measure its thermodynamic properties. That said, MD is not yet fast enough to reproduce processes of actual interest for chemists or biologists, such as, protein-protein or protein-ligand interactions, or dynamics of large multidomain proteins.
Multiple shorter MD runs can get you a longer simulation time faster (because these shorter trajectories can be run simultaneously). However, you need to have some prior understanding about the process of interest, for a smarter choice of starting configurations. Strategies like Markov State Model (see ref. below) can then be applied to extract useful kinetic and thermodynamic metrics out of those simulations.
Pande, V. S., Beauchamp, K., and Bowman, G. R. (2010) Everything you wanted to know about Markov State Models but were afraid to ask. Methods 52, 99-105
Multiple, the more the better, "sufficiently" long MD simulations is a must for one to be able to assess the statistical (in)significance of the geometric and energetic data.
My personal experience is that what matters (free energy) is never statistically significant: MD simulation is not a tool for quantitative predictions. After 12 years of running MD simulations i am converging (no pun intended) to the conclusion that MD simulation is nothing else than a fancy way to waste a lot of time and tax-payers' money . . .
These days, for example, i am trying to figure out what to do with free energy differences of 1 kcal/mol with standard deviations of 2 to 5 kcal/mol.
- Badges
- Science topic