Without knowing your research purpose, it is difficult to define "different initial conformation". For example, if you are studying the interaction between the substrates and an enzyme, "initial conformation" means the relative position between them. Usually, repeating the MD simulation run from the beginning is enough, since Gromacs will generate the random initial velocity (don't use the same seed ). The reviewer may want to check if the results are obtained by chance or they are general or universal.

As Chenlin Lu mentioned, mostly the multi copy simulations are needed just to ensure reproducibility or significance of your data. Yet it is not ensured that all your copies will converge to same data with shorter runs (run time lesser than convergence time ) as different starting points (different initial conformers or fresh starts with different seeds) may sample different regions of your phase space attainable at root temperature (T in your simulation). However, more number of relatively smaller runs is expected to reproduce at least significance of your observed conformational changes. And again, it is not possible to suggest different initial conformations without knowing your 'research purpose'.

Thank you for the answers. I have done the MD simulation in order to see the aggregation behaviour of protein in presence of osmolytes. In that case do I need to do multiple simulation using the same initial conformation or different by changing the random initial velocity.

You can do multiple run with same initial conformation with different velocities. But I think it will be wise to do multiple runs with different placements of the proteins/peptides (whatever you are using) with respect to each other as you are studying their aggregation and if random placements of the proteins does reproduce the aggregation behaviour, it will validate your piece of hypothesis (whatever it is).

Thanks Souvik. If i have single protein with 150 AA how do i write different conformations.

You said you are studying aggregation behaviour of protein. So, how can you do that with a single protein? May be it's not what I think you are doing. Whatever, if you have just a single protein in your system, then what I said earlier is not of any use. In that case, fresh starts with new velocity assignments is possibly your best choice for multiple runs.

Thanks Souvik for the clarification. I have one more question. How to check that there is no artifact in the simulation? As i have checked it by the calculation of minimum distance of the protein to its periodic image and its value is found to be >2 nm. I have done the simulation of protein in triclinic box with box size of 7*16*6 nm^3. Also the cut -off values are

r_list= 1.2

rvdW= 1.2

r_colomb = 1.2

I don't know what kind of artifacts you are talking about. If your system is well equilibrated and quiet stable structurally at a long run without any unusual conformational change (by visual inspection), I think its fine. Do you think the minimum image distance should not be > 2nm? If you have maintained a solvent slab of ~1nm at all sides of the protein, the minimum distance should be ~2nm I think.

Thank you Souvik for the answers.