Designing coarse grained-and atom based-potentials for protein-protein docking

Background: Protein-protein docking is a challenging computational problem in functional genomics, particularly when one or both proteins undergo conformational change(s) upon binding. The major challenge is to define a scoring function soft enough to tolerate these changes and specific enough to distinguish between near-native and "misdocked" conformations. Results: Using a linear programming (LP) technique, we developed two types of potentials: (i) Side chain-based and (ii) Heavy atom-based. To achieve this we considered a set of 161 transient complexes and generated a large set of putative docked structures (decoys), based on a shape complementarity criterion, for each complex. The demand on the potentials was to yield, for the native (correctly docked) structure, a potential energy lower than those of any of the non-native (misdocked) structures. We show that the heavy atom-based potentials were able to comply with this requirement but not the side chain-based one. Thus, despite the smaller number of parameters, the capability of heavy atom-based potentials to discriminate between native and "misdocked" conformations is improved relative to those of the side chain-based potentials. The performance of the atom-based potentials was evaluated by a jackknife test on a set of 50 complexes taken from the Zdock2.3 decoys set. Conclusions: Our results show that, using the LP approach, we were able to train our potentials using a dataset of transient complexes only the newly developed potentials outperform three other known potentials in this test.