Molecular Docking Based On Multi-objective Differential Evolution

With time passing by and the progress of technology, a growing number of threedimensional information of bimolecular are being determined. To deal with suchconformational information has become a challenge for bioinformatics. Moleculardocking is an in silicon tool that aims to predict the binding mode when two or moremolecules interact. Molecular docking is widely used in modern structure based drugdesign (SBDD), such as predicting the binding mode and binding energy betweensmall molecular and its receptor as well as molecular virtual screening (VS).Commonly, molecular docking involves two parts: the conformation searchingalgorithm and the scoring function. Conformation searching algorithm is to find themost likely binding mode between donor molecular and its receptor around thebinding site. At least six dimension searching space is involved in conformationsearching for rigid docking, including three translation freedom and rotation freedom.More variables are needed for flexible ligand and protein docking. Different strategiesare available for conformation searching at present, such as molecular mechanicsmodeling, geometry matching and stochastic algorithms. To develop an efficient andreliable conformation searching algorithm is still a challange for molecular docking.The scoring function, which evaluates the binding modes of docking is anotherimportant issue in molecular docking. The scoring function should evaluate thebinding mode correctly and discriminate different binding modes. There are mainlythree types of scoring functions: the force field based scoring function, the empiricalscoring function and the knowledge based scoring function. The most commonly usedscoring function is empirical based, which involves the evaluation of differentinteraction components in docking and the empirical weights for differentcomponents. We present a multiple objective differential evolution molecular dockingalgorithm. Unlike the traditional scoring functions commonly get one sum value frommultiple weighted molecular interaction components, we treat different interactioncomponents as multiple optimization objects. Due to the lack of precise computationof interaction energy and the involvement of empirical weights, the one sum scoringfunction may not precisely reflect the binding energy. In our algorithm, the interactioncomponents include the van der Waal interaction, hydrogen bonding, electrostatic andhydrophobic. The docking process optimize multiple interactions rather then theirweighted summation.Due to the huge complexity of conformation searching in molecular docking, thesearching algorithms used in molecular docking must be efficient and avoid non-localsearching. Stochastic algorithms, such as genetic algorithm, simulated annealing andMonte Carlo algorithm have been used and proved their efficiency in moleculardocking. In this paper we applied the differential evolution algorithm in moleculardocking. The differential evolution is a simple and efficient optimization algorithm,which encode with real number and evolute with differential vector. Previoussimulation result has shown that the differential evolution is superior to otherevolution algorithms in efficiency and accuracy.A large number of molecular ensembles will be generated during moleculardocking. Due to the special distribution of docking ensembles that commonly liesaround the molecular surface, these structural data are commonly uniformlydistributed, thus the exterior structure of the ensemble may represent essentialdocking mode. Using clustering algorithm to analysis these structures will improvethe docking process. We present a novel geometrical clustering algorithm and appliedit on molecular docking process. The algorithm is a top-down hierarchical process,with exterior nodes as seed nodes for classification. The selection of seed points is bysearching the nodes that comprise the largest tetrahedron. Compared to otheralgorithms, the geometrical clustering algorithms classified the data more uniformly. The right binding modes are more likely to be extracted than other binding modes.With the geometrical clustering algorithm, the docking process is more precise andefficient.