I am trying to build a tool for molecular docking of antibodies and antigens. This is for a student project, and as I don't know a lot about immunology, I wanted to check with you all what do you think of the approach I am using, and advices on what could be added / improved / removed.
Inputs : two PDB files, one containing informations of the antibody and the other one, the corresponding antigen Output : a dozens PDB files containing the best configurations found.
1 - Prediction of antibody's CDRs, via proABC
2 - Measure of circular variance of CDR's residues, to have an idea of the 3D shape of the CDR (are they external or internal residues?)
3 - On the structure of the antigen : calculate the residues depth, and keep only surface residues (accessible to solvant).
4 - On the structure of the antigen : check circular variance of the whole surface. If the circular variance shows a shape that could be complementary to the CDRs one, then we store this region.
5 - Use energy fields to determine which regions have the lower energy when putting it in contact with CDRs.
6 - Use statistical potential to improve the specificity.
7 - Measure RMSD of the interesting regions to check for eventual superimposition (after alignement of these regions).
From steps 4 to 7 I use a score ( arbitrary score ), and I keep only the dozens configurations that have the best scores. The program first does those steps with all 3 CDR, then only with pairs of 2, and then with only one. Is this correct ? I couldn't find a better way to look for the paratope region. Is there a better way that you could think of ?