Full Optimization of Linear Parameters of a United Residue Protein Potential

TL;DR

This paper presents a method to optimize all linear parameters of the UNRES protein potential to reliably predict native-like conformations as low energy states across multiple proteins, enhancing the accuracy of protein structure prediction.

Contribution

It introduces a comprehensive protocol for optimizing all linear parameters of the UNRES potential, improving native conformation prediction without extra multi-body terms.

Findings

Optimized parameters successfully predict native-like conformations as global minima.

Method works for multiple proteins simultaneously.

Native conformations are energetically favored after optimization.

Abstract

We apply the general protocol of parameter optimization (Lee, J. et al. Phys. Chem. B 2001, 105, 7291) to the UNRES potential. In contrast to the earlier works where only the relative weights of various interaction terms were optimized, we optimize all linear parameters of the potential. The method exploits the high efficiency of the conformal space annealing method in finding distinct low energy conformations. For a given training set of proteins, the parameters are modified to make the native-like conformations energetically more favorable than the non-native ones. Linear approximation is used to estimate the energy change due to the parameter modification. The parameter change is followed by local energy reminimization and new conformational searches to find the energies of native-like and non-native local minima of the energy function with new parameters. These steps are repeated until the potential predicts a native-like conformation as one of the low energy conformations for each protein in the training set. We consider a training set of crambin (PDB ID 1ejg), 1fsd, and the 10-55 residue fragment of staphylococcal protein A (PDB ID 1bdd). As the first check for the feasibility of our protocol, we optimize the parameters separately for these proteins and find an optimal set of parameters for each of them. Next we apply the method simultaneously to these three proteins. By refining all linear parameters, we obtain an optimal set of parameters from which the native-like conformations of the all three proteins are retrieved as the global minima, without introducing additional multi-body energy terms.