Chem-Bio Informatics Journal Vol. 17(2017)

CYP3A4 contributes to the metabolism of more than 30% of drugs in clinical use. Predicting the sites of metabolism (SOM) by CYP3A4, as well as the binding modes, for target compounds is important for the design of metabolically more stable drugs. Precisely predicting the structures of CYP3A4–ligand complexes is enormously challenging owing to the high number of conformational possibilities with its numerous binding substrates. We previously described a method for predicting the SOM of carbamazepine by means of docking and molecular dynamics (MD) simulations starting from multiple initial structures. To validate our method, we have now applied it to tolterodine, which is more flexible than carbamazepine. In addition, we evaluated the effectiveness of two methods for selecting the initial structures for MD. In analyzing the MD trajectories, we calculated the frequency with which carbon atoms at each of four groups of the tolterodine molecule approached to within a certain cutoff distance of the heme iron, and we also calculated binding free energy. We found that compared to the other three groups, the position to the experimentally determined SOM was close to the heme most frequently and had the lowest average ∆Gbinding. For selecting the MD initial structures, clustering on the basis of protein–ligand interaction fingerprints (PLIF) was substantially more robust at predicting accessibility compared with clustering based on root-mean-square deviations. These findings demonstrate that our method is applicable for a flexible ligand and that PLIF clustering is a promising method for selecting structures for MD. We succeeded to predict the experimentally determined SOM of tolterodine together with the appropriate binding mode. The predicted binding mode is useful to design metabolically more stable compounds.

Several researchers have focused on the inference of genetic networks as a process for extracting useful information from gene expression data. Their work has led to the proposal of a number of methods for genetic network inference. Yet the genetic networks inferred by these methods often contain large numbers of false-positive regulations along with the true-positives. One effective way to reduce the number of erroneous regulations is to apply inference methods that use a priori knowledge on the properties of the genetic networks. The existing inference methods adopting this approach generally use a priori knowledge and the observed gene expression data simultaneously to determine whether or not the target genetic network actually contains each of the candidate regulations. In this study, we establish a new framework for “using a priori knowledge after genetic network inference.” The framework uses a priori knowledge only to modify the genetic network that has already been inferred by the other inference method. Based on this framework, we propose a new inference method that uses multiple kinds of a priori knowledge about genetic networks. The proposed method effectively combines multiple kinds of knowledge and computes the confidence values of regulations. Here, we confirm the effectiveness of the proposed method by applying it to artificial and actual genetic network inference problems. While only a small improvement is gained from the use of multiple kinds of a priori knowledge, we can improve the performance of many other existing inference methods by combining them with the method we propose here.