Journal of Physical Chemistry B, Vol.108, No.27, 9421-9438, 2004
Parametrization of backbone-electrostatic and multibody contributions to the UNRES force field for protein-structure prediction from ab initio energy surfaces of model systems
The multibody terms pertaining to the correlation between backbone-local and backbone-electrostatic interactions in the UNRES force field for energy-based protein-structure prediction, developed in our laboratory, were reparametrized on the basis of the results of high-level ab initio calculations on relevant model systems. MP2/6-31G(d,p) ab initio calculations were carried out to evaluate the energy surfaces of pairs consisting of N-acetyl-N'-methylacetamide molecules (AcNHMe, which model a regular peptide group) and N-acetyl-N',N'-dimethylacetamide molecules (AcNMe2, Which model a peptide group preceding proline) at various intermolecular distances and orientations. For each pair, the calculated ab initio energy surface was subsequently fitted by a sum of Coulombic and Lennard-Jones components. Then, the restricted free-energy (RFE) surfaces of pairs of free peptide groups as well as the RFE factors corresponding to the coupling of backbone-local and backbone-electrostatic interactions in model tetrapeptides were calculated by numerical integration, with the use of the ab initio-fitted simplified energy functions and the ab initio energy maps of model terminally blocked amino acid residues calculated recently (Oldziej, S.; Kozlowska, U.; Liwo, A.; Scheraga, H. A. J. Phys. Chem. B, in press, 2003). Next, analytical expressions based on Kubo's generalized cumulant theory from our previous work were fitted to the resulting RFE surfaces to parametrize the backbone-electrostatic and multibody terms in the UNRES force field. The computed coefficients of the cumulant-based expressions are different from those derived earlier, which had been based on the ECEPP/3 force field. To complete the force-field parametrization, the weights of the energy terms were determined, and the coefficients of the cumulant-based expressions were refined simultaneously by using our recently developed method of hierarchical optimization of a protein energy landscape using the protein 1IGD. The resulting force field was able to predict significant portions of the structures of proteins with alpha, beta as well as both alpha and beta structure correctly.