| Computational prediction of protein-protein interactions is crucial for a better understanding of processes such as metabolic control, signal transduction, and gene regulation, whereas the ability to dock small ligands to proteins is the key to rational drug design strategies. We have developed three algorithms that are useful in different stages of protein-protein and protein-small molecular docking. First, a general clustering algorithm, based on the biophysical knowledge of interactions, was developed for the analysis of docking decoys. The method was applied to the problem of ranking protein-protein and protein-small ligand complexes generated by rigid body docking methods, and was shown to perform comparably with other available scoring functions. The second algorithm developed is a novel global search method based on the Fast Fourier Transform (FFT) correlation technique. The method can score billions of complex conformations using pairwise knowledge based potentials in reasonable time. This algorithm was implemented in the PIPER program for protein-protein docking. The program showed improved performance in comparison with other existing methods when applied to the docking of the protein pairs in the well known protein docking benchmark, and was successfully used in the CAPRI (Critical Assessment of PRediction of Interactions) experiment. The third algorithm builds on the results of the clustering approach described above. The clusters correspond to local minima on the free energy surface of the protein-protein complex, and include both near-native states and false positives, i.e., conformations that have low energy but are far from the native structure. In order to discriminate between near-native and non-native states, we have studied the stability of the region corresponding to each cluster. The basic idea of the method is to resample each region with a detailed energy function using Monte Carlo Minimization (MCM), and check whether the majority of the MCM trajectories stay in the region and form a funnel (i.e., a stable region of attraction), or dissipate, indicating a non-stable local minimum. Discarding non-stable clusters for the complexes in the protein docking benchmark reduces the number of false positives by 50%, at the same time preserving the ones close to the native state. |
