ABGEN - A KNOWLEDGE-BASED AUTOMATED APPROACH FOR ANTIBODY STRUCTURE MODELING

Immunoglobulin (Ig) amino acid sequences are highly conserved and ofte n have sequence homology ranging from 70 to 95%. Antigen binding fragm ents (Fab), variable region fragments (Fv), and single chain Fv (scFv) of more than 50 myeloma proteins and monoclonal antibodies (mAb) have been crystallized and display a high degree of structural similarity. Based on this observation, several homology modeling approaches have been developed for the prediction of Fab and Fv structures prior to th eir experimental determination. We have extracted features from existi ng Ig sequences, 44 known Fab and Fv structures to create an automated AntiBody structure GENeration (ABGEN) algorithm for obtaining structu ral models of antibody fragments. ABGEN utilizes a homology based scaf folding technique, and includes the use of invariant and strictly cons erved residues, structural motifs of known Fab, canonical features of hypervariable loops, torsional constraints for residue replacements an d key inter-residue interactions. The validity of the ABGEN algorithm has been tested using a five-fold cross validation with the existing F ab structures. Molecular mechanics and dynamics methods have been impl emented with ABGEN models to accurately predict two Fab structures of anti-sweetener antibodies prior to crystallographic determinations.