| As one of the common forms of neurodegenerative disease, Alzheimerdisease (AD) can seriously alter patients’ memory and cognitive function. ADcannot be cured and is degenerative. The oldest hypotheses exist trying toexplain the cause of the disease, on which most currently available drugtherapies are based, is the cholinergic hypothesis. It postulates that the reducedsynthesis of the neurotransmitter acetylcholine contributes substantially to thecognitive decline observed in those with advanced age and AD. Thus,acetylcholinesterase, which catalyzes the hydrolysis of acetylcholine andserves to terminate synaptic transmission with very high catalytic activity, haslong been considered as a target for AD therapy and acetylcholinesteraseinhibitor tacrine was approved as the first agent for the treatment of AD. Thiswork built several classification models for acetylcholinesterase inhibitors andnon-inhibitors, and QSAR (Quantitative Structure Activity Relationships)models for predicting the bioactivity of each inhibitor.In the first part, several models were built using a Support VectorMachine for classification of acetylcholinesterase inhibitors and non-inhibitors. 721inhibitors and3892decoys were collected. After data preprocessing,finally714inhibitors and1983decoys were left. Each molecule was initiallyrepresented by211ADRIANA.Code and334MOE descriptors, which wereused separately and combined together for modeling. Correlation analysis andStepwise regression, F-score and attribute selection methods in Weka wereused to find the best reduced set of descriptors, respectively. Additionally,models were built using a Support Vector Machine and evaluated by5-,10-fold and leave-one-out cross-validation. The best model gave a MatthewsCorrelation Coefficient (MCC) of0.99and a prediction accuracy (Q) of99.66%for the test set. The best model also gave good result on an externaltest set of86compounds (Q=96.51%, MCC=0.93). The descriptors selectedby our models suggest that H-bond and hydrophobicity interactions areimportant for the classification of AChEIs and decoys and it is also confirmedby a docking simulation.In second part, Several QSAR (Quantitative Structure ActivityRelationships) models for predicting the inhibitory activity of404Acetylcholinesterase inhibitors were developed.19global descriptors,8shapedescriptors and1024RDF descriptors were calculated using ADRIANA.Code.The whole dataset was split into a training set and a test set randomly or usinga Kohonen’s self-organizing map.16descriptors were selected by Correlationanalysis and Stepwise regression method. Then the inhibitory activity of404Acetylcholinesterase inhibitors was predicted using four models based on Multilinear Regression (MLR) analysis and Support Vector Machine (SVM)methods, respectively. For the test sets, correlation coefficients of all ourmodels over0.90were achieved. Y-randomization test was employed toensure the robustness of our models and a docking simulation was used toconfirm the descriptors we used.We hope that the our work focusing on buliding of classification andquantitative prediction models for acetylcholinesterase inhibitors can be usefulfor further studies on virtual screening and structure–activity relationship ofacetylcholinesterase inhibitors. |
