Vol.2 No.1 2009

75/88

Research paper : Advanced in-silico drug screening to achieve high hit ratio (Y. Fukunishi et al.)−72−Synthesiology - English edition Vol.2 No.1 (2009) We applied the same scheme on tautomers. Accordingly, although the compound configurations are still not chemically strict, the ion forms of our compound database are more reliable than typical open software such as babel and openbabel, which frequently generate high-energy tautomers.6 Prediction of non-selective compoundsQuestion and comment (Takatsugu Hirokawa)As application examples of protein-compound affinity matrix discussed in section 4.11, non-selective or low-selective compounds (or highly selective compounds), which nonselectively bind to many target proteins, may possibly be predicted by using the database. If so, this may realize highly unique annotations such as in-silico frequent hitters or in-silico chemical alerts for target selectivity. Please discuss about the possibility of predicting non-selective compounds based on the database, although this might have been conducted already.Answer (Yoshifumi Fukunishi)It is an insightful question. Frequent hitters account for several tens of percent in VS and thus increase the cost and are the bottleneck of screening processes. Approximately 20 % of the predicted compounds are frequent hitters in our screenings. Recently, we are collecting several dozens of compounds that are considered the frequent hitters from literatures (J. Med. Chem. 2003, vol. 46, page 4477-4486, J. Med. Chem. 2002, vol45, page 137-142) and developing their 3D structures. Once the data have been prepared and included in the computation of protein-compound affinity matrix, we may be able to find a property that contributes to a high score against any target protein. However, there is a report that most frequent hitters form micelle colloids when observed under an electron microscope, and thus, the adsorption of the micelle to proteins could be a cause of the “frequent hit” (J. Med. Chem. 2002, vol. 45, page 1712-1722). If the frequent hitters exhibit non-selectivity against target proteins as unimolecular compounds, it will be possible to distinguish frequent hitters by docking simulations. However, if the micelle formation is the cause of the “frequent hit,” docking simulations, which assume infinite dilution condition, will not be able to discriminate frequent hitters. Thus far, as a result of applying a solubility prediction based on molecular descriptors to analyze the aqueous solubility of frequent hitters, it was found that highly hydrophobic molecules tended to be frequent hitters and thus could be distinguished from drug molecules that are not frequent hitters (P1-06 at the Chem-Bio Informatics Society (CBI) Annual Meeting 2008 International Symposium). If the water solubility of a compound determines its likeliness to be a frequent hitter, the micelle formation should be the primary cause of “frequent hit.” Nevertheless, there remains a possibility that docking simulations could be more effective to predict frequent hitters. Although it will take time, we will continue to characterize frequent hitters.Although we analyzed the side effects caused by the non-selectivity of compounds by MTS and DSI methods, thus far, no clear association between the side effects and non-selectivity of compounds was found. COX2, a typical target of nonsteroid anti-inflammatory drug (NSAID), and COX1, functioning as a gastric mucosal protector, are the enzymes with 60 % homology in their amino acid sequences; thus, it was an issue of concern that NSAID could cause gastric ulcers by inhibiting not only COX2 but also COX1. Recently, COX2 selective NSAIDs (e.g. coxibs) have been developed. Even though we then examined whether selective and non-selective NSAIDs can be distinguished by using protein-compound affinity matrix, it was not successful. In fact, the COX2 selectivity of coxibs is relatively low; at a concentration of 80 % inhibition of COX2, selective and non-selective NSAIDs inhibited approximately 20 % and 80 %, respectively, of COX1 activity (Proc. Natl. Acad. Sci. USA. 1999, vol. 96, page 7563-7568). Hence, the drug selectivity, in this case, is not a black-and-white property but a matter of degree. We consider it possible to distinguish highly selective and non-selective compounds by using protein-compound affinity matrix and currently prepare the structures of approximately 1500 proteins for docking simulations. Although an actual analysis cannot be performed due to the limitation of computational capacity, it will be possible anytime soon.

※このページを正しく表示するにはFlashPlayer9以上が必要です