Vol.2 No.1 2009

70/88

Research paper : Advanced in-silico drug screening to achieve high hit ratio (Y. Fukunishi et al.)−67−Synthesiology - English edition Vol.2 No.1 (2009) reconstructed and the 3D structure was regenerated.4.7 Identification of enantiomers and generation of isomersIf each of four chemical bonds of an atom such as carbon binds to different molecular fragments, the central atom will be the chiral center. Hence, discrimination of the four molecular fragments bound to the central atom will be necessary. In the method that we have developed, the bonds to the central atom are cut off and the molecular fragments are identified through comparison using a method similar to the algorithm described in section 3.2. Although the process will be slightly more complex if the central atom is part of a ring structure, a similar discrimination method is used. If only a single chiral center exists in a molecule, the enantiomers can be generated by converting the coordinate of each atom, (X, Y, Z), to (X, Y, −Z). If two or more chiral centers exist in a molecule, the bonds need to be reconnected; we used confgeneC, a newly developed software, for this purpose.4.8 Computation of atomic charges by quantum chemical calculationsIn quantum chemical calculations, electron spins and charges of molecules are necessary in addition to the molecular structures. Molecules used for drug development should not be radicals and also rarely exhibit magnetic properties; hence, the molecules were assumed to be closed-shell with zero spin. We developed an algorithm that automatically computes the molecular charge that stabilizes the system from the information on chemical bonds. The charge of the whole molecule is considered to be the sum of the formal charge of each atom. For example, the formal charge of a carbon atom is considered to be zero if the sum of the chemical bonds is four and +1 if it is three; the charge of nitrogen is +1 if the sum of the chemical bonds is four and zero if it is three; and the charge of oxygen is zero if the sum of the chemical bonds is two and −1 if it is one. The molecular charges were then evaluated by summing the formal charges obtained in this manner from the whole molecule.There are several methods of computing atomic charges. The Gasteiger method[10] assigns electron affinity to each atom and evaluates the equilibrium electron distribution where atoms pull their electrons with each other based on the organic electron theory. A rough estimation requires less than a second for most molecules. In semi-empirical quantum chemical calculations, the AM1 and PM3 models (recently, PM7) of MOPAC[11] are well-known. The PM3 model is an excellent method in which an effective Hamiltonian is derived by fitting parameters so that the heat of formation can be represented; however, structures commonly observed in pharmaceuticals such as an amide bond cannot be accurately computed. The AM1 model also evaluates an effective Hamiltonian by fitting parameters; although it is inaccurate in predicting heat of formation, most structures such as an amide bond can be calculated correctly. However, this method occasionally fails to accurately predict the atomic charges of ring structures that contain nitrogen atoms. If the molecular structure is defined, the computation time will normally be several to several dozen seconds and is approximately proportional to N3 for the size of an atom, N. The computational accuracy of the charge is very high. In ab-initio calculations of quantum chemistry, wave functions and partial atomic charges are generally computed by the RHF/6-31G* and restrained electrostatic potential (RESP) methods[12], respectively. Although this approach evaluates charges extremely accurately, the computation time will normally be several to several dozen minutes if the molecular structure is defined and is proportional to N4 for the size of an atom, N.Atomic charges will be meaningless unless protein-compound bindings are accurately computed. Thus, docking calculations of 132 protein-compound complexes were performed by sievgene[13], our protein-compound docking simulation software*term1. As a result, accurate structures were obtained with a probability of 56 % by RHF/6-31G* (with an accuracy of 2 Å), with 2–3 % lower probability by MOPAC AM1, and with about 5 % lower probability by the Gasteiger method. A small-scale drug screening experiment using approximately 10,000 compounds was also performed targeting several proteins such as cyclooxygenase-2 (COX-2) and thermolysin. It was found that the hit rate was higher if the molecular charges were more accurate, and the hit rate was only a few percent lower even when the Gasteiger method was applied.Since the atomic charges of several million molecules will be calculated, the computation time should be prioritized. In addition, it was found to be unnecessary to use a method that is as accurate as RHF/6-31G*. Hence, we decided to employ the MOPAC AM1 method for the computation of charges because the compound DB will be the overall foundation. Although MOPAC generally requires a MOPAC-specific input format, we modified it so that we can input and output the mol2 file format, which is a standard format to represent compounds in the field of drug discovery. For this purpose, we are distributing a patch file to modify MOPAC free of charge.4.9 Determination of equivalent atomsThe charge of three protons in a methyl group should be configured to be chemically equivalent. The determination of atomic equivalency is necessary for computing atomic charges.The equivalency of arbitrary atoms i and j is considered to be as follows: If i = j, the atoms are obviously equivalent. However, if this is not the case, and if atoms i and j do not directly bind to each other, all of the atoms binding to atom i should be equivalent to those binding to atom j, whereas if atoms i and j do bind, all of the other binding atoms should be equivalent to each other.

※このページを正しく表示するにはFlashPlayer9以上が必要です