Perform Replicas #research #scicomm #moleculardynamics #compchem

Ajouté : 2026-06-05

[00:00:00]Hi. So this video is dedicated particularly to researchers. And Most of Us Have Been Doing MD Simulations Wrong.

[00:00:05]This is not a clickbait video. It's a genuine video and you can read this paper that I am linking down to get more insight. Please remember MD simulations are also experiments. And whenever you do an experiment, you do replicas in biology or in any other field. You do let's say three ah expe experiments and then take out a standard deviation and have a mean and then report that. Similarly in MD simulations you also need to run replicas. Conclusion on one MD simulation is disasterous. Ok? This paper again that I have linked down. Please go ahead, read it. It's open access. So when you do an MD simulation you basically ah generate random velocities. When you run, you do the minimization and then you do the equilibration and then you do the production.

[00:00:45]In the minimization there are no velocities that are equal. But after that step, they are random velocities that are assigned.

[00:00:52]And these random velocities can take the MD simulation in any trajectory. Now understand this, that for very small molecules, yes, if you do a one- microsecond simulation, there is a chance that you might be able to cover its entire confirmation. Then, if you are mapping its entire energy landscape, this is fine. But for biomolecular simulations, particularly simulations with protons, and with, you know, thousands of atoms. It's almost impossible for a one- microsecond simulation, even one millisecond simulation, to map its entire energy landscape. So, that means when you assign these random velocities, it basically pushes it, it can kind of bias the simulation and pushes it into a particular direction. So, at least, if you do three simulations with three random velocities, and if still the simulation moves in a particular way, or if the biomolecule behaves in a particular way, then you can be confident. Statistically confident that yes, this is what is happening, and in this paper, we give an example of hydrogen bonding: if you run one simulation, there's a good chance that you've only got five hydrogen bonds, and if you run another simulation, there's a good chance that you've got even hydrogen bonds, but the actual average is eight hydrogen bonds. So, if, for example, a scientist is doing a study in which they are looking at how many hydrogen bonds are being formed between two proteins, and in one case they find it's five, and in the other case they find it's 11, they will draw completely different conclusions. One wood says that the number of hydrogen bonds is reducing. So, for example, if the average number of hydrogen bonds is eight, and you, one researcher, runs MD simulation and finds that the number of hydrogen bonds is five. So they will reduce that the number of hydrogen bonds decreases. Where as another researcher finds that there are 11 hydrogen bonds. They will reduce that the hydrogen bond increases because of which whatever is whatever and because of which they conclude something which is completely incorrect because the actual average is eight.

[00:02:23]Right? So this is it I am talking about.

[00:02:26]Computer simulations are experiments in the end and you need to treat them as experiments. I like I said I have been guilty of it. And almost everyone around me who is even a hard core a simulation person and of course experiment I don't expect much from experiment list. The Typical Wet Lab Experiment List. But even people who are directly working in MD simulations are also unaware of this. Desimulation is so democratized. So even if you're not performing MD similarities, you know someone else who is performing MD similarities, or it's being used in your research article as a supporting study, ask for replicates and do share it with everyone around you.