Receptor guided QSAR studies of S-trityl-L-Cysteine analogues as potent Eg5 inhibitors.

kinesin Eg5, a member of the kinesin-5 superfamily, is responsible for the formation and maintenance of bipolar spindle in the early prometaphase stage of mitosis. Since Eg5 plays a crucial role in cell division, inhibition of this protein causes mitotic arrest, which can lead to cell death. In the past few years, a large number of chemicals were synthesized and evaluated as Eg5 inhibitors [1, 2]. Kosielski and coworkers introduced S-trityl-L-cysteine (STLC) as a potent Eg5 inhibitor. In the recent years computational drug design methods have gained significant contribution to the development and optimization of drugs, which may shorten the cycle of the drug development [3]. Among the theoretical methods, quantitative structure–activity relationship (QSAR) is a well-established technique to quantitative correlate the physicochemical properties of molecules to their biological properties. In the present report, we performed a hybrid docking-QSAR approach to obtain a statistical QSAR model to the study of the structural features of STLC analogues synthesized by Wang et al. that influenced in Eg5 inhibitory activity [4]. In the first step, in order to consider the conformational change of STLC analogues in Eg5 binding site the favorable conformation of STLC analogues in the binding site of Eg5 was obtained from the docking methods. Then the structural molecular descriptors of these optimized conformers were computed to developed statistical QSAR models. Multiple linear regression was applied to choose the best subset of descriptors. The selected descriptors explain that the gravitation index [all pairs], and H attached to C1 (sp3)/C0 (sp2) play important roles in Eg5 inhibition. Multiple linear regression (MLR), square support vector machine (SVM), and artificial neural network (ANN) were used as QSAR models, respectively. The results of this model reveals the superiority of SVM models over two others models. The squared correlation coefficient (R2), root mean square error (RMSE) and standard error (SE) for SVM are: R2=0.964, RMSE=0.16 and SE=0.150 for the training set, and R2=0.920, RMSE=0.287 and SE=0.25 for the test set. Leave one out cross validation test was used for assessment of the predictive power and validity of models which led to cross-validation correlation coefficient (Q2) of 0.941 and standardized predicted relative error sum of squares (SPRESS) of 0.225 for SVM model.