For using Autodock, if I would do the following (quick and dirty protocol):

1. Do a cleanup of the PDB file, removing unwanted hetero-atoms, ions etc. that appear due to various additives and salts in the crystallization buffer.

2. Check for any missing amino-acids and protein segments, and build them using Modeller, if possible.

3. Run a molecular dynamics simulation (about 10 ns or so) of the protein in water to enable the protein to get relaxed, especially the segments are built in the previous step. Usually what you obtain is quite close to refined. I suppose, you can also use the links you mentioned to get a refined structure at this stage.

I hope that this is helpful.

But since you mentioned that you want to perform a docking study (I presume, against a drug library) and I keep getting these kind of queries (mostly biologists who often use these docking software's as black-box) , I want to take this opportunity to put the following on record.

Fact: The protocol above will give you a refined structure, but that does not mean it is appropriate for a docking study.

Of course, Autodock will run properly and will give you very nice looking docked structures, but more often than not the results do not have any practical significance. Proteins are highly dynamic molecules and an effective docking study needs to take into consideration all the important conformational states of the protein, rather than just a single snapshot obtained from PDB database. In Autodock, you can mention regions that should be treated flexibly, but it certainly does not generate all the possible protein conformations.

A proper docking study should first generate all the possible (stable and metastable) states, and use these multitude of conformations for docking. Currently, there are efforts to use enhanced sampling algorithms (within the MD framework) for the same, but most of these require a lot of computational power and time.

In conclusion, using autodock on a single PDB structure amounts to nothing. I have heard from colleagues that some of the paid softwares are much superior in this aspect, but I don't have a first hand experience using them. Getting all the relevant states first is the key. Thereafter, I would say any docking software would work quite effectively.

Abhishek Acharya thank you very much.

what would be MD method which consume the least amount of computational power ?