In this particular case there were several potential templates. The article by Chothia et al., that we refer to shows one (good) example of template choice. We decided to go for another template, because the one we took has the same overal fold, but one of the loops was longer so that upon modlling we would also get that loop. That loop was/is so important because it contains some of the residues for which mutation information is available.
It is a common misunderstanding that homology modelling is not possible below 25% sequence identity. The truth is that below 25% identity the homology between template and model cannot be detected by classical sequence alignment techniques. If, however, the homology can be detected by other means, and an alignment can be made, then one can also build a model. The quality of the model will be low when one analyses the fine details.
Therefore, you are strongly discouraged to look at the model in detail. At the atomic level there are even some bad errors left (funny bond lengths, funny angles, etc.). We deliberately decided to leave that because for the present study the relative positions of side chains was the major topic. We know that we got that as good as could be done given all information available in the summer of 1997. Cleaning up those smaller errors by energy minimisation or molecular dynamics techniques would create the risk that certain residues are moved around so much that they no longer are located near where we expect them from the analysis of all available data.
So, bare with us, but this is a model that can only be used for a qualitative interpretation of the mutation information, and that is all what we did in the article.
Please address modelling questions to "Vriend@EMBL-Heidelberg.DE"