Hello everyone, I'm having an issue with a review of a paper.
I performed 50ns MD simulations on docking poses to know if the predicted poses were stable.
Among the literature doing similar things I found one of the analysis I could do on the obtained data was FEL on the first two Principal Components (PC1 and PC2), then counting the quantity and area of the more energetically favorable regions in the graph, and comparing that between the unbound protein, and protein in complex with different ligands, so I proceded to do it.
Now, reviewer 2 states that 50ns is too small sampling time so we should remove the FEL analysis. I want to know how can I answer to that critique, as if based in literature I can find papers doing such an analysis in that sampling time, but I don't want to start arguing that if I don't know for sure if the sampling time is enough to conclude something about the most probable metastable conformations of the protein in each condition (non-bound or the different complexes).
Thanks in advance, any thoughts on the matter will be of great help.
It depends on the complexity of your system, if it is not very dynamic, maybe your 50 ns simulation time is fine otherwise your dynamics has to be extended. One way to respond the reviewers is to extend your simulation times and compare your results by 50 and for example 200 ns to demostrate that your system reaches stability during the first 50 ns.
How many data points do you have -- millions? How many independent MD simulations did you run on the system -- at least three? How did you assess the uncertainty of the quantitative results -- and is the uncertainty low enough not to render the quantitative results insignificant?
It really varies from system to system. I also think 50ns is really small for protein free energy calculation. I usually do it for 500ns. You can check your RMSD distribution over time.
Whether 50 ns is enough or not is on-its-own a mute point: just a few years ago, 500 ps was enough. Just that we have better computers does not mean that running longer MD simulations provides better data.
i have executed well over 1000 MD simulations; the longest were 10 ns. It is better to run multiple short MD simulations than one long. Data for free energy analysis should be collected about every 10 MD steps, so if the time step is 2 fs, one can collect in this way 2.5 million energies from one 50-ns MD simulation.
Do not expect a true convergence in free energies obtained from MD simulations; absent a sheer luck, it is a feat impossible to achieve. And this also means that if the data are interpretable, rejoice for being so lucky, and interpret the data, regardless of how they were obtained. Anything goes. Embrace serendipity.
And hope that the anonymous 'experts' in the field (i mean, the reviewers) have a bit of a common sense.
The length of simulation for convergence depends on the system under study. For simple systems such as dihedral flipping, several ns is enough. However, for complex systems, several microseconds or even longer simulation times are required. If you are only doing PCA analysis, actually 50 ns is acceptable but it is indeed a bit short with modern computational resources. I think increasing it to 500 ns could be a nice choice. However, you should know that modern simulation papers published often starts with at least several micro-second simulations and then do some analyses.
- Badges
- Science topic
- Similar topics
- Chemistry
- Theoretical Chemistry
- Computational Chemistry
- Molecular Dynamics