Knowledge-based grouping of modeled HLA peptide complexes

Human leukocyte antigens are the most. polymorphic of human genes and multi ple sequence alignment shows that such polymorphisms are clustered in the f unctional peptide binding domains. Because of such polymorphism among the p eptide binding residues, the prediction of peptides that bind to specific H LA molecules is very difficult. In recent years two different types of comp uter based prediction methods have been developed and both the methods have their own advantages and disadvantages. The nonavailability of allele spec ific binding data restricts the use of knowledge-based prediction methods f or a wide range of HLA alleles. Alternatively, the modeling scheme appears to be a promising predictive tool for the selection of peptides that bind t o specific HLA molecules. The scoring of the modeled HLA-peptide complexes is a major concern. The use of knowledge based rules (van der Waals clashes and solvent exposed hydrophobic residues) to distinguish binders from nonb inders is applied in the present study. The rules based on (1) number of ob served atomic clashes between the modeled peptide and the HLA structure, an d (2) number of solvent exposed hydrophobic residues on the modeled peptide effectively discriminate experimentally known binders from poor/nonbinders . Solved crystal complexes show no vdW Clash (vdWC) in 95% cases and no sol vent, exposed hydrophobic peptide residues (SEHPR) were seen in 86% cases. In our attempt to compare experimental binding data with the predicted scor es by this scoring scheme, 77% of the peptides are correctly grouped as goo d binders with a sensitivity of 71%. Human Immunology 61, 460-466 (2000). ( C) American Society for Histocompatibility and Immunogenetics, 2000. Publis hed by Elsevier Science Inc.