Prediction of peptide binding to MHC molecules via molecular modeling and global optimization

Development and application of molecular modeling and global optimization methods have led to successful prediction of features of peptide binding to the MHC molecule HLA-DRB1*0101. Because peptide binding to MHC molecules is essential to the immune response, the development of such methods for understanding and predicting the forces that drive this binding is crucial for pharmaceutical design and disease treatment. Underlying the development of these prediction methods are two hypotheses. The first is that pockets formed by the peptide binding groove of HLA-DRB1*0101 are independent, permitting decomposition of the prediction problem into the prediction of peptide amino acids binding in individual pockets and the prediction of peptide amino acids binding between these pockets. The second hypothesis is that the native state of a peptide bound to a protein corresponds to the system's lowest free energy. This hypothesis permits formulation of the prediction problem as the construction and minimization of descriptions of the system's free energy. These hypotheses have shaped the development of a three-stage peptide binding prediction method. The first, or anchoring, stage involves atomistic level modeling, deterministic global optimization, and three methods of solvation: solvent-accessible area, solvent-accessible volume, and Poisson-Boltzmann electrostatics. This stage predicts binding affinities of peptide amino acids for pockets of HLA-DRB1*0101 by determining computationally an amino acid's global minimum energy conformation. The second, or designing, stage involves distance-based force field modeling and mixed-integer optimization. This stage predicts which amino acids are preferred in peptide positions by determining their global minimum energy of interaction both with HLA-DRB1*0101 and with a T-cell receptor. The third, or binding, stage involves atomistic level modeling, solvation with Poisson-Boltzmann electrostatics, local minimization, and incorporation of the predictions of the anchoring and designing stages. This stage predicts the binding affinity of entire peptides for the HLA-DRB1*0101 molecule. Computational results from these peptide binding prediction methods are in agreement with x-ray crystallography data and experimental binding assays. These results demonstrate that molecular modeling and global optimization are a promising approach to peptide binding prediction.