A carefully built model of the Fv regions of the source antibody and
in some cases the humanised antibody is a common denominator in
many of the successful humanisation experiments. Nowadays, it is
possible to build a model completely automatically using commercial
packages or academic servers
However, the danger of allowing a computer to make all the decisions is
highlighted in the humanisation of antibody
AT13/5 where the interaction between residues
at positions 29 and 78 (Kabat numbering) in the heavy chain was not
modelled correctly. The experience of an expert in protein structure
modelling is always welcomed.
Additionally, in the author's own humanisation experience,
it is better to also model the constant regions of the antibody, since
interactions at the variable/constant domain interface are likely to affect the
avidity of the molecule.