Theoretical Studies On The Predictions Of P-glycoprotein Inhibitors And Substrates

ATP-binding cassette （ABC） transporter protein family is a large class of transmembrane proteins, which are expressed in many organs. ABC transporter protein protects the cells from toxic substances by pumping the toxic substances out of the cells with the energy provided by ATP. P-glycoprotein （P-gp） is the most representative member of the ABC transporter proteins superfamily. P-gp is a pseudo-symmetrical heterodimer with each monomer consisting of two bundles of six transmembrane （TM） and two nucleotide-binding domains （NBDs）. The binding pockets of P-gp are mainly located at the TM areas. The function of NBD is to bind ATP, thus providing energy for P-gp substrate binding. It is reported that P-gp can transport many chemically and structurally unrelated drugs and agents, resulting in the MDR phenomenon that accounts for chemotherapeutic failure in the treatment of cancers. besides, P-gp is highly involved in absorption, distribution, metabolism and elimination （ADME） of drugs. Thus, researches on P-gp inhibitors and substrates have important significance for cancer treatment.The thesis is divided into two parts. In the first part, we firstly reported an extensive database of1273molecules that are categorized into P-gp inhibitors and non-inhibitors. The impact of eight important physicochemical properties on P-gp inhibition was examined. Then, the decision trees were built from a training set of973compounds using the recursive partitioning （RP） technique and validated by a test set of300compounds. The best decision tree correctly predicted83.5%of the inhibitors and67.0%of the non-inhibitors in the test set. Finally, naive Bayesian categorization modeling was applied to establish classifiers for P-gp inhibitors. The Bayesian classifier gave average correct prediction for81.7%of973compounds in the training set using a leave-one-out cross-validation procedure and81.2%of300compounds in the test set. We found that the inclusion of molecular fingerprints could improve the prediction significantly. Moreover, as an unsupervised learner without tuning parameters, the Bayesian classifier employing fingerprints highlights the important structural fragments favorable or unfavorable for P-gp transport, which offers extra valuable information for designing new efficient P-gp inhibitors early in the drug discovery process.In the second part, we reported an extensive P-gp substrates dataset of822compounds which collected from previous literatures. Based on the large dataset, we checked the impact of various important physicochemical properties on P-gp substrates activity. Based on P-gp inhibitors dataset reported by our group, a dataset of735P-gp inhibitors and422substrates were assembled. We examined the differences of P-gp inhibitors and substrates on the basis of this dataset. We employed the docking method to study the differences between the binding features of P-gp substrates and inhibitors. Then, the Bayesian classifiers were built from a training set of622compounds and validated by a test set of200compounds. The naive Bayesian classifier based on molecular properties and the ECFC₁0fingerprints yielded77.8%accuracy for the training set using the leave-one-out （LOO） cross-validation procedure and83%for the test of200compounds. Moreover, as an unsupervised learner without tuning parameters, the Bayesian classifier employing fingerprints highlights the important structural fragments favorable or unfavorable for P-gp transport. Finally, we discussed the reasons of misclassification for some molecules.