Novel Modelling-optimization Approach and Monte Carlo Method on QSAR Study of Bortezomib Drugs

Multiple linear regression (MLR) as modeling tool and Imperialist Competitive Algorithm (ICA) as optimization techniques employed to choose the best set of descriptors and The CORAL software has been used as a tool for linear prediction of -log( IC50) (empirical negative logarithm of half of maximal inhibitory concentration) for Bortezomib derivatives. A high predictive ability was observed for the MLR-ICA model with the best number of empires/ imperialists (nEmp) 90 with root-mean-sum-square errors (RMSE) of 0.0121 and correlation coefficient (R2predict) of 0.9896 in gas phase.
The 25 data sets were randomly splitted into the training set, the calibration set, the test set in the Monte Carlo method and the number of compounds in the each set (n), correlation coefficient (R2) , cross-validated correlation coefficient (Q2), standard error(s) were calculated 13, 0.9826, 0.9780, 0.161 in training set; and n=6, R2= 0.8463 , Q2=0.7377, s=0.715 in test set in the Threshold (T) of 2 and probe of 3, respectively.
From the MLR-ICA method, it was revealed that Espm15u, R5p+, B06 [O-O], F03[N-N], F07[C-O], MATs3m, RDF125v are the most important descriptors. From Monte Carlo simulations, it was found that the presence of double bond and ring, absence of halogens are the most important molecular features affecting the biological activity of the drug.
It was concluded that simultaneous utilization of MLR-ICA and Monte Carlo method can lead to a more comprehensive understanding of the relation between physico-chemical, structural or theoretical molecular descriptors of drugs to their biological activities and facilitate designing of new drugs.