Beta-Secretase Inhibitors QSAR Study

Beta-secretase (BACE1, membrane-associated aspartic protease 2, beta-site APP cleaving enzyme 1, aspartyl protease 2and beta-site amyloid precursor protein cleaving enzyme 1) is a member of aspartic protease family, which is an aspartic-acid protease important in the formation of myelin sheaths in peripheral nerve cells. In 1990s, hardy raised ?-amyloid peptide (A?) as the core of A? doctrine, abnormal hydrolysis metabolism of amyloid precursor protein, which A? cause the formation of neurofibrillary tangle and senile plaques of aberrantly folded proteins in the brain. It leads to Alzheimer's disease. BACE1 is an important target for the treatment of AD. Computer-aided Drug Design is a kind of descreasing cost methods that is used into develop new drug. Qualitative classification and quantitative prediction models are widely used in new drug reseach. In this study, we build qualitative classification model and quantitative prediction models, in order to explore BACE1 inhibitors structure-activity relationship.(1) Self-Organization Map (SOM) and Support Vector Machine method were used to build qualitative classification models of BACE1 inhibitors.294 BACE1 inhibitors,5?M as the threshold of activity inhibitors and low activity inhibitors, were split into training set and test set by Self-Organizing Map (SOM) and Random. Use ADRIANA.Code software to calculate descriptors of BACE1 inhibitors. Then, methods, Pearson correlation analysis plus stepwise regression, F-score and SVMAttributeEval of Weka, were used to screen descriptors. SOM and SVM method were used to establish classification model. The obtained model showed a good predictive accuracy of over 82.61% on the test set. Model2D was the optimum classification model in this work. The classification accuracy rate of test set was 89.02%. The AUC of test set was 0.86, and Y-randomization validation showed a steady Model2D. Extended Connectivity Fingerprints (ECFP4) was conducted to explore the relation between BACE1 inhibitors space structure and bioactivity. Finally, we found that six substructures were likely to affect the bioactivity of BACE1 inhibitors. They could be used to guide and optimize the prodrug of BACE1 inhibitors.(2) Multiple Linear Regression and SVM method were used to build quantitative prediction models of BACE1 inhibitors.294 BACE1 inhibitors were split into training set and test set by Self-Organizing Map (SOM) and Random. We used ADRIANA.Code software to calculate descriptors of BACE1 inhibitors. Pearson correlation analysis and stepwise regression were used to screen descriptors. Multiple Linear Regression and SVM method were used to establish models. The models of training set which was generated by SOM were better than that was generated by random method. For modeling method, SVM method was more effective than MLR method. The robust model Model3G showed squared correlation coefficient (R2) of training set up to 0.95. The test set showed R2=0.74. Mode3G was steady, testing by Y-randomization method.A comprehensive descriptor analysis of the models told us that the atom structure information, hydrogen, molecular charge, electric potential and atomic polarization were important to the bioactivity of BACE1 inhibitors.In this study, the classification models and quantitative prediction models, which we built, performed a good forecasting ability.