We propose an automated protocol for designing the energy landscape of a protein energy function
 by optimizing its parameters. The parameters are optimized so that
 not only the global minimum energy conformation becomes native-like, but also the conformations distinct from the
 native structure have higher energies than those close to the native one.
We classify low-energy conformations into three groups,
super-native, native-like, and non-native conformations.
 The super-native conformations have all backbone dihedral angles fixed to their native values,
 and only their side-chains are minimized with respect to energy.
 On the other hand, the native-like and non-native conformations all correspond to the
 local minima of the energy function.
 These conformations are ranked according to their root-mean-square deviation (RMSD) of backbone coordinates
 from the native structure, and a fixed number of conformations with the smallest RMSD values are selected
 as native-like conformations, whereas the rest are considered as
non-native ones.
 We define two energy gaps  Eg1 and Eg2.
 The energy gap Eg1 ( Eg2 ) is the energy difference between
 the lowest energy of the non-native conformations and the highest energy of the native-like (super-native) ones.
The parameters are modified to decrease both Eg1 and Eg2. In addition, the non-native conformations
 with larger values of RMSD are made to have higher energies relative to those with smaller RMSD values.
We successfully apply our protocol to the parameter optimization of the UNRES potential energy,
using the training set of betanova, 1fsd,
 the 36-residue subdomain of chicken villin headpiece (PDB ID 1vii), and the
10-55 residue fragment of staphylococcal protein A (PDB ID 1bdd).
The new protocol of the parameter optimization shows better
performance than earlier methods where only the difference between
the lowest energies of native-like and non-native conformations
was adjusted without considering various degrees of
native-likeness of the conformations. We also perform jackknife
tests on other proteins not included in the training set and
obtain promising results. The results suggest that the parameters
we obtained using the training set of the four proteins are
transferable to other proteins to some extent.