The efficient screening of lead compounds or drug candidates for efficacy and safety is critically important during the early stage of drug development. Compounds are usually screened from a diverse 'chemical space' based only on its pharmacological effects, but this screening is not enough to guarantee drug safety. To solve this problem, we devised a chemical space that takes into account interaction information with proteins such as drug transporters. We also created and evaluated a mathematical model for predicting compound-transporter interactions. This was achieved by first generating an interaction correlation matrix based on drug transporters and their corresponding inhibitor compounds. To implement a screening scheme that takes into account interaction with drug transporters, we created a model using Canonical Correlation Analysis (CCA) that makes use of the known information on interaction between drug transporters and their corresponding inhibitors. Cross-validation of the model gave satisfactory test results and a physiologically relevant chemical space was constructed based on the model.

View full abstract

Proteins with a charge periodicity of 28 residues (PCP28) were found recently in the human proteome, and many of the annotated PCP28 were located in the nucleus (Ke et al., Jpn. J. Appl. Phys. 2007). The physical properties of the amino acid sequences were analyzed to detect the difference in the physicochemistry between the nuclear and cytoplasmic PCP28 and develop a software system to classify the two types of PCP28. A significant difference in the global parameters from the entire sequence and the local parameters around a segment with the highest positive charge density was found between the nuclear and cytoplasmic PCP28. The global classification score included the densities of proline and cysteine, and the negative charge density, while the local score included the symmetry of the charge distribution, the density of cysteine, and the positive charge density. A prediction system was developed using the global and local scores, which possessed a sensitivity and specificity of 92% and 88%, respectively. The mechanism of translocation of proteins to the nucleus is discussed using the parameters relevant to the predictive system.

View full abstract