“Soft docking”: Matching of molecular surface cubes

Abstract

Molecular recognition is achieved through the complementarity of molecular surface structures and energetics with, most commonly, associated minor conformational changes. This complementarity can take many forms: charge-charge interaction, hydrogen bonding, van der Waals' interaction, and the size and shape of surfaces. We describe a method that exploits these features to predict the sites of interactions between two cognate molecules given their three-dimensional structures. We have developed a “cube representation” of molecular surface and volume which enables us not only to design a simple algorithm for a six-dimensional search but also to allow implicitly the effects of the conformational changes caused by complex formation. The present molecular docking procedure may be divided into two stages. The first is the selection of a population of complexes by geometric “soft docking”, in which surface structures of two interacting molecules are matched with each other, allowing minor conformational changes implicitly, on the basis of complementarity in size and shape, close packing, and the absence of steric hindrance. The second is a screening process to identify a subpopulation with many favorable energetic interactions between the buried surface areas. Once the size of the subpopulation is small, one may further screen to find the correct complex based on other criteria or constraints obtained from biochemical, genetic, and theoretical studies, including visual inspection. We have tested the present method in two ways. First is a control test in which we docked the components of a molecular complex of known crystal structure available in the Protein Data Bank (PDB). Two molecular complexes were used: (1) a ternary complex of dihydrofolate reductase, NADPH and methotrexate (3DFR in PDB) and (2) a binary complex of trypsin and trypsin inhibitor (2PTC in PDB). The components of each complex were taken apart at an arbitrary relative orientation and then docked together again. The results show that the geometric docking alone is sufficient to determine the correct docking solutions in these ideal cases, and that the cube representation of the molecules does not degrade the docking process in the search for the correct solution. The second is the more realistic experiment in which we docked the crystal structures of uncomplexed molecules and then compared the structures of docked complexes with the crystal structures of the corresponding complexes. This is to test the capability of our method in accommodating the effects of the conformational changes in the binding sites of the molecules in docking. For this, an uncomplexed trypsin inhibitor (4PTI in PDB) and an uncomplexed trypsin (3PTN in PDB) were used in one case, and an uncomplexed lysozyme (1LYZ in PDB) and the antibody portion of a lysozyme-antibody complex (2HFL in PDB) in the other (the crystal structure of this antibody alone is not known). Our results verify the importance of both the geometric and energetic complimentarity in docking. In both cases the correct solutions were found among the top 500 solutions (out of over 10,000 to 30,000 solutions) saved from the first stage of geometric docking alone. These solutions were then screened in the second stage of the method, based on favorable and unfavorable energetic interactions. Only a small number of solutions showed overall favorable interactions and the correct docking solution was among them (within the top 4 solutions for 4PTI and 3PTN, and within the top 12 solutions for 1LYZ and 2HFL variable domain). Thus, our “soft docking” procedure was able to reduce drastically the search space resulting in a small number of candidate solutions while tolerating the conformational changes associated with complexing processes in these two tested examples.