An automated approach to modelling class II MHC alleles and predicting peptide binding

We present an automated method for constructing 3D models of class II MHC structures that uses constraint logic programming to select side-chain conformations. The resulting models are used by a "peptide threading" program that attempts to predict peptides from a protein sequence that will bind strongly to particular MHC alleles. This method follows a comparative modelling approach in basing the model structures on experimentally determined MHC-peptide structures. However, constraints are used to ease open the peptide binding groove so that the modelled MHC structure is a less specific fit for the co-crystallised peptide in the starting structure. Preliminary results indicate that MHC models that have been constructed in this way enable the peptide threading program to make binding predictions that are comparable with those obtained when using experimentally determined MHC structures