Computer-aided Of Structure-activity Relationships(SAR)Study Of Cyclooxygenase(COX)Inhibitors

Cyclooxygenase?COX?is a key rate-limiting enzyme in the arachidonic acid?AA?anti-inflammatory network system.It can catalyze AA into prostaglandin H2?PGH2?,which causes a series of inflammatory reactions,such as fever,pain,etc.There are two major subtypes of COX,which is COX-1 and COX-2.These two subtypes are also targets for traditional nonsteroidal anti-inflammatory drugs?NSAIDs?.This work aims to establish a comprehensive COX inhibitors database,which containing the structure,activity and selectivity information.Based on the database,we systematically carried out the structure-activity relationship?SAR?studies of COX-1 inhibitors and selective COX inhibitors.This work are mainly focus on the following aspects:?1?Qualitative classification?i.e.structure-activity relationship?SAR??study and scaffolds classification study of cyclooxygenase-1?COX-1?inhibitors.The collected 1530 COX-1 inhibitors were divided into highly active and weakly active inhibitors using the IC50 value of 10 ?M as the threshold.Two splitting methods,splitting randomly and by Kohonen's self-organizing map?SOM?method,were used to divide the COX-1 inhibitors into training and test sets.Two descriptors,descriptors representing physicochemical properties?CORINA descriptor?and molecular fingerprints representing the features of structure fragments?MACCS fingerprints?,were used as two inputs.Three machine learning algorithms,support vector machine?S VM?,decision tree?DT?and random forest?RF?,were used to establish twelve qualitative classification models based on COX-1 inhibitors.The best model had an Matthew correlation coefficient?MCC?of 0.93 for the test set.In order to analyze the relationship between the structure and activity of COX-1 inhibitors,we manually divided the 1530 COX-1 inhibitors into nine subsets according to their different scaffolds and summarized the structural features of each subsets.?2?Quantitative structure-activity relationship?QSAR?study of cyclooxygenase-1?COX-1?inhibitors.The structure and activity values?IC50 values?of 181 COX-1 inhibitors whose IC50 were definite and measured by enzyme immunoassay were collected.These 181 COX-1 inhibitors were then randomly divided into training and test sets three times,and nine filtered CORINA descriptors were used as an input.Six QSAR models were built by multiple linear regression?MLR?algorithm and support vector machine?S VM?algorithm.These regression models can predict the specific activity values of compounds according to their structures.The correlation coefficient?R?on the training and test sets of the best model were 0.91 and 0.86,respectively.The mean square error?MSE?on the training and test sets were 0.18 and 0.21,respectively.?3?The structure-activity relationship study of selective cyclooxygenase?COX?inhibitors.We first established a database of COX selective inhibitors containing 686 cyclooxygenase inhibitors.According to their inhibition ratios of COX-1 and COX-2,they were classified into selective COX-1 inhibitors and selective COX-2 inhibitors.Then these 686 selective COX inhibitors were divided into training and test sets randomly and by Kohonen's SOM methods.The filterd CORINA descriptors,MACCS fingerprints and ECFP₄ fingerprints of selective COX inhibitors were used as an input.There were eighteen classification models on selective COX inhibitors were established by support vector machine?SVM?,decision tree?DT?and random forest?RF?methods.The prediction accuracies of these models for training and test sets were greater than 94%.The MCC values of these models were greater than 0.66,which proves to be statistically significant.