Actually I understood what is going wrong... modeller renumbers
everything so at first it finds the residues of chain A that are
missing... However, the next ones will be 128-130 and not 1-3...
Is there a way to keep the initial numbering and not let modeller
renumber my structures?
In this case you can't use option (1) because you won't have a PDB file
containing the missing residues and their numbers (unless you manually
create a dummy one; remember that Modeller only parses the ATOM/HETATM
records, not SEQRES). (2) might be easiest if the numbering is sequential.