You are pulling along the centre of mass of the two chains and it is therefore physically very likely for something like what you describe to happen. You need to do the umbrella sampling simulations to be able to identify which conformation is relevant. It may be a pure artefact of the initial pulling and would therefore disappear once the umbrella sampling simulations have converged.

Dear Akhil,

It seems you've made a mistake in the preparation of an index file. Some part of the red protein (several amino acids) is in the index file of blue protein. I think this might be a reason that part of the red protein is not moving during the pulling simulations due to the restrained potential. If you have further questions just let me know.

Kind regards,

Just a methodological note for your project. If you're only interested in the thermodynamic effect of a mutation and are worried about spurious intermediates affecting your numbers, you might want to try the alchemical approach instead, where you only need to consider the two (presumably better characterized) endpoints, the dimer and the monomer.

There's a lot of resources on that these days, including a recent "best practices" preprint (Preprint Best Practices for Alchemical Free Energy Calculations

Thank you Ioana Mariuca Ilie, for pointing me in right direction. I am doing umbrella sampling simulations for each frame (COM distance 0.1). Yes i m worried about PMF after the umbrella simulation. Is this artefact going to affect the my final results.

Dear Jamoliddin Razzokov, i dont think so I have made mistake in index file because when i m replicating my simulation (with same parameters) few time i have observed smooth pulling of chain B. We have given restrained only on chain A.

Dear Miłosz Wieczór, thank you for suggesting me better alternative and methodology. I am interested in how mutations (most of the mutant are on surface) affect the dimerization of chain A and chain B. so first we have introduce the single mutation on chain A and B and perform MD for 200 ns than trying to pull the chain B from chain B than plot PMF to see the energy profile and compare with the mutant type.

Akhil Kumar I get that, a while ago I tried a similar approach dealing with protein-DNA complexes, but what I found out was that free energy profiles for dimerization are very hard to converge to a level where you can pick up a small effect (1-2 kcal/mol) from the noise. In principle, you would have to have enough sampling in each window to visit *all* the accessible intermediate states with good statistics, so you either put a ton of resources or get numbers that are very far from reliable.

You can try and see for yourself, but ultimately an alchemical approach will be more reliable, at least for dimerization (not to say 100% reliable, as sampling & force field issues still matter, but more than with PMFs). It's not always preferable (for example, you don't learn anything about the intermediates, and hence about free energy barriers/kinetics), but I'd say it's the way to go if you're just interested in getting your ΔΔG for a complex process with a poorly specified pathway.

Regarding the remark by Jamoliddin, it's exactly my point - in your pathway the protein doesn't properly dissociate, one part remains bound while the red domain partially unfolds, and finding meaningful dissociation pathways is inherently hard. You can solve these technical issues in many ways, using minimal distance or a combination of the number of native contacts and COM distance as the reaction coordinate, but each variant will have its drawbacks. However, if you want to stick to the PMF approach, I'd suggest using some variant of the minimal distance approach and introducing orientational restraints to limit the number of non-specific intermediates - for two protein variants these will mostly cancel out, improving the "signal-to-noise" ratio in the difference between the profiles.