Selection of chain depends on the presence of active site. Moreover, you can select whole protein only if the active site is not known.

If you are doing blind docking you have to take the whole protein. Now, most of the crystal structures in PDB is given as the asymmetric unit (unit cell). One such unit cell may contain multiple chain of the same protein, but that does not mean that the protein is a multimeric protein. A protein might be a monomer but appear as a multimer in a particular crystal structure. In such as case, you have to take one chain only, as that is the physiological state of the protein. Similarly, a multimeric protein may appear as a monomer in an asymmetric unit. In that case, you have to build the appropriate multimer from the monomer using symmetry (or look for the "biological assembly") and then do the docking. For more details go to the following link:

https://pdb101.rcsb.org/learn/guide-to-understanding-pdb-data/biological-assemblies

First and foremost, one is supposed to check the physiological state of the protein prior to docking/MDS. Subsequently, one can decide which chain to keep and which to discard. There is no harm in keeping the entire protein chains while performing docking/MDS, except the computational cost. One more thing, better mimicking of the biological molecules can be done by performing MDS.

There are no specific rules how to select the chain for docking and in most cases it does not matter because the chains are usually very similar. Thus, normally just the first chain or any chain is taken for no reason. However, in some cases chain selection might influence the docking results. Therefore, when selecting the chain there are some things that should be considered:

1) some loops or amino acid residue side chains might not be resolved in some of the chains, but resolved in the others

2) loops and amino acid residue side chains might have different conformations

3) in some cases even binding mode of the ligand might differ

4) different chains might have different water molecules (other solvent molecules, ions etc.) resolved

5) some chains might have better resolution than the others

6) finally, you might have a crystal structure of a biologically relevant multichain complex of several same or different molecular objects in which case it might be relevant to take the whole complex

What Uttam Pal has said is what you need to keep in mind. If a protein entry in PDB has multiple chains, then it doesn't mean it is biologically a multimeric protein. It could just be an artefact of the crytallisation process. In case of actual homomeric proteins also, if each homomer is independently a biologically functional unit, you can use just one chain for docking studies. All you need is an intact binding/active site.

That brings us to the choice of chain in these scenarios. Even though for a homomeric protein both chains are supposed to be identical, in case of PDB structures it may not always be the case. Often, the electron density of both the monomeric chains is not resolved similarly. One chain may be resolved better than the other, and thus one of the chains might have a few residues, especially in terminal regions or loops, with missing coordinates or poorly resolved electron density. Thus you should choose the chain which represents a more intact structure, and most importantly, has all the residues of the biologically relevant binding/active site.