If there are multiple structures are available, you may find one in which the loop is resolved, either due to better resolution or because its conformation is constrained due to crystal packing.

If the missing loops are short enough, homology modeling programs may patch in the loop themselves (of course, you have to give them the sequence including the missing residues). If that is not the case, you can use a loop modeling server, e.g. https://modbase.compbio.ucsf.edu/modloop/ , http://rcd.chaconlab.org ,

http://opig.stats.ox.ac.uk/webapps/fread/php/, http://bioserv.rpbs.univ-paris-diderot.fr/services/DaReUS-Loop/ (see http://www.vls3d.com/index.php/links/bioinformatics/3d-structure-prediction/modeling-loops for a more complete overview).

In some cases, the loop is not missing because of missing electron density, but because it has been replaced by a shorter loop in the crystallization construct to improve crystallization. Check the publication associated with the pdb code of your structure. I would not run the structure as is, since a chain break introduces a pair of charges (new N- and C-term) that may alter the electrostatics of your system, eg. altering the stability of adjacent helixes by interacting with the helix dipole, or altering ligand interaction.

I would say, use gromacs to add residues to your structure. Sometimes when crystals are created there are additions residues added to it which are not the part of a protein. One has to add these manually using gromacs then run the molecular dynamics. If one runs MD without these RESIDUES and due to high photon energy of the pulse there is a lot of ionization, one will not be able to see the affect of these residues to the final charge states.

If the incomplete region is far from the region of interest (and not expected to measurably affect the region of interest): exclude the incomplete region from simulation by placing it outside the non-periodic simulation sphere.

You can add those missing residues before going for MD simulation. You can use Swiss PDB viewer, which adds automatically and you can use other repositories also. It is an necessary step before MD otherwise toplogies may have defects.

Scrwl can also be used for side chain building: http://dunbrack.fccc.edu/SCWRL3.php/

Caution: Adding residues to the C-terminus is easy, Swiss PDB viewer also allows addition to the N-terminus. To build loops is more difficult, you have to also achieve loop closure. You do this either by scanning a loop database (matching the main-chain coordinates of the flanking regions on either side of the loop and the length of the loop segment to loop segments in a database of loops extracted from the pdb e.g. in the Swiss PDB viewer. The loop can also be modeled de novo using an algorithm that searches for conformations that will achieve loop closure (e.g. in rosetta). Both approaches get increasingly more difficult and their results less reliable, the longer the missing loop segment is.

If the residues are at the N or C terminals and they are not crucial for your system, leave them as it is. If the missing residues are in-between and they are just a few residues, you can use Chimera (Modeller at the background) to complete them.

You can use any modeling programs to fill in the missing residues (Swiss Model or MODELLER or Sequence module in Chimera). However, if your missing residues are far away from the site of interest then you can just exclude it.

First of all you can check the location of missing residues, if it is in N/C-terminal and functionally not required (that you can only find out through literature's) than leave it. Otherwise you have to build in both the cases like, if it is functionally required or present in-between the protein.

Now the question is how many residues are missing and how to model the structure. If the residues number are 1-4 or 6 you can add through

1. Swiss Model server

2. Discovery Studio

3. Scrwl

4. Sybyl

5. FALC and many more tools

But if the missing residues are more than that then kindly switch to homology modelling. Before that perform pdb blast and try to find out suitable template against your structure.