Yes, 10 ns may be enough provided that (1) the purpose of the MD simulations is to calculate the binding free energy, (2) no large-scale conformational changes are expected that would affect the stability of the protein-ligand complex, (3) multiple MD trajectories are generated starting from different initial conditions (seed for initial velocities), and (4) protein-ligand and protein-solvent interaction energies for the free energy calculations are stored every 10 MD simulation steps.

Frankly, no. Usually several times of 20 - 50 ns (excluding equilibration time) are tolerable, but not by every journal.

If you are considering publishing the results then "NO". With high impact journals, at least 100ns is the trend. But, 50ns is also acceptable in some journals.

Yes, 10 ns may be enough provided that (1) the purpose of the MD simulations is to calculate the binding free energy, (2) no large-scale conformational changes are expected that would affect the stability of the protein-ligand complex, (3) multiple MD trajectories are generated starting from different initial conditions (seed for initial velocities), and (4) protein-ligand and protein-solvent interaction energies for the free energy calculations are stored every 10 MD simulation steps.

The simulation time should be consistent with the time scale of the actual structural phenomenon you are about to study. Residence time for a ligand in a protein pocket ranges from microseconds to hours.

, , , And to thoroughly sample a microsecond timescale, the total simulation time must be at least three orders of magnitude longer. The same would apply to the hours, which will turn into weeks of simulation time, about 3600000000000000000 1-fs timesteps. Not very feasible. Better to take care of the initial structure and the force field parameters, and run multiple short MD simulations to sample the configurational space close to the initial structure -- and for this 5-ns simulations may be enough: if the difference in the binding free energies between the first and second half of the simulation varies by less than 2 kcal/mol, one can call it a "convergence."

Actually, I think you must continue simulation until the system is converged. Sometimes, 10ns is enough sometimes not.

Alireza Mohebbi, may i ask, what is your measure of convergence?

Dear Martin Klvana some parameters, including RMSD and Rg fluctuation are used for obtaining convergence. When the system reached to a consistence level of RMSD or Rg or PCA (for e.g. after 2ns) your system is converged. By considering the time point at which system is converged you can do subsequent statistical analyzes. I hope this comment be helpful.

Thank you everyone for your feedback, it really helped me gain a better insight. As of now I have downloaded the results for 10 ns but I have also started a dynamic production job for the complexes for another 90ns.

Ishita Shreshtha, Look how your system converges in the timescale. For my system, I did a 25 ns production run. When I see the RMSD profiles, I can clearly make, that my system is reaching convergence in the last 10 ns, i.e. (from 15ns -25ns), based on the fluctuations or noise, depending on the statistics. Thereafter I perform metadynamics calculations.

Hope it helps.

Regards,

Deb

Debashish Banerjee Thank you for your suggestion, will definitely try this

10 ns will only equilibrate the system.

Jitendra Kuldeep, no amount of simulation time will equilibrate controlled explosion.

Martin Klvana I think the context of the question is with regard to finding a likely drug candidate molecule for a target.

Jitendra Kuldeep, it is possible to obtain protein-ligand binding free energies with sub-2 kcal/mol uncertainty by running a set of single-digit nanosecond MD simulations; depending on the system and simulation setup, the equilibration phase may be as short as 200 ps; of course, it depends on what kinds of phenomena are involved in the protein-ligand interaction: while hydrogen-bonding and stacking interactions can be described very accurately with sub-nanosecond MD simulations, large-scale conformational changes will never be thoroughly sampled; either way, there is no such thing as a "minimum simulation time to obtain useful results"; MD simulation is, first and foremost, an idea generator, and a priori assumption that no idea may be generated within 1 ps, not to mention 10 ns, MD simulation, is arbitrary, and as such may not be always useful.