i would calculate average RMSDs (and standard deviations) and i would plot the RMSDs as histograms (to see how much the two RMSDs overlap), i.e., i would consider the snapshots of each system as belonging to an ensemble rather than to a timeline.

Hi Keong,

It seems that your systems reach equilibrium before 25 ns (black) and 40 ns (red). Although you would get better results after performing at least three simulations for each system and performing a cluster analysis on the equilibrium region to analyze how the interactions between the ligand and protein changes in the equilibrium phase.

Hi Keong,

It seems that your systems reach equilibrium before 25 ns (black) and 40 ns (red). Although you would get better results after performing at least three simulations for each system and performing a cluster analysis on the equilibrium region to analyze how the interactions between the ligand and protein changes in the equilibrium phase.

As for the RMSD value in your simulation, the exact value is not very meaningful unless the simulation is long enough. For you case, only 100 ns simulation is performed. With modern computational resrouces, several hundered ns simulations could be performed within several days. No statistically meaningful value is obtained in short unbiased simulations. I recommend you run several micro-seconds before calculating these quantities.

RMSD is not necessarily lower than some threshold like 0.3 nm. The crystal structure is deposited in different condition compared with the biological environment. Therefore, the crystal structure should, in principle, differ from the solvated protein-ligand system. However, we expect the difference to be not significant and thus use the crystal structure as the starting structure to simulate these systems.

Only monitoring RMSD is not very informative. I recommend you to calculate some structrual properties such as secondary structure, RMSF, and so on.

Zhaoxi Sun , is it possible to obtain 'statistically meaningful data' from molecular dynamics simulations?

i would rather run many short (few nanoseconds) MD simulations than one long (microseconds, milliseconds, seconds . . .).

@Martin,

Could you please tell me how to calculate average RMSDs (and standard deviations) and how to plot the RMSDs as histograms.

Puneet Kumar, let L be a Python list of RMSD values:

import numpy

import matplotlib.pyplot as plt

# average (avg) & standard deviation (sd)

avg = numpy.average(L)

sd = numpy.std(L)

# histogram

plt.hist(L)

I don't think it is necessary to go for microseconds molecular dynamics production run for studying the stability of a ligand/drug in the pocket of the protein. For complex molecular dynamics, you don't go that far to study stability. You only need to go as far as microseconds when you are studying the folding of a protein.