DARS (Decoys As the Reference State) potentials

The PIPER program was used with a new class of structure-based potentials called DARS (Decoys As the Reference State), based on the inverse Boltzmann approach (Kozakov et al., 2006). A statistical potential between two atoms ai and aj of types I and J, respectively, and located within a certain cutoff distance D, is defined by ε_IJ = -RT ln (p_IJ), where R is the gas constant, T is the temperature, and pIJ denotes the probability of two atoms of types I and J interacting. This probability is approximated by the frequency p_IJ = ν_obs(I,J) / ν_ref(I,J) where ν_obs(I,J) is the observed number of interacting atom pairs of types I and J, and ν_ref(I,J) is the expected number of interacting atom pairs of types I and J assuming an appropriate reference state. While ν_obs(I,J) can be directly determined by counting the number of intermolecular interactions between atoms of types I and J in a database of protein-protein complexes, the selection of the reference state remains a critical feature. The general assumption in the reference state is that the specific interactions determining the distribution of interaction sites are removed as much as possible. The most frequently used reference state is defined by ν_ref (I,J) = ν_obs X_I X_J where ν_obs is the total number of interacting pairs with the distance constraints d < r_i,j < D, and X_I and X_J are the mole fractions of atom types I and J, respectively.

The novelty of the DARS approach is that we generate a large decoy set of docked conformations using only shape complementarity as the scoring function (i.e., without any account for the atom types), and use the frequency of interacting atom pairs in these decoy structures as the reference state. Thus, developing the potential we compare the frequency of contacts between two specific atom types in X-ray structures of protein complexes to the frequency of contacts in the decoys that are devoid of specific interactions. Since the goal is finding complex conformations close to the native among the many structures that all have good shape complementarity, this scoring scheme is very natural, as it rewards the occurrence in the interface of the atom pairs that are frequently seen to interact in the native complexes. Due to the use of the more accurate DARS potential in PIPER, we do not need further filtering, and the top 1000 structures are retained for clustering.

Although the current version of the DARS potential improves docking results for almost all complexes, it works best for enzyme-inhibitor. We are currently developing special DARS potential for antigen-antibody complexes.