Development of new metal-thiosemicarbazone complexes using visual screening methods and in silico models

The stability constants (logb11) of forty-two new metal-thiosemicarbazone complexes were predicted based on the results of the quantitative structure-property relationship (QSPR). The QSPR models were developed from 88 logb11 values of experimental complexes by using the multivariate linear regression (QSPRMLR) and artificial neural network (QSPRANN). Four descriptors such as xch9, xv0, core-core repulsion and cosmo area were found out in the best of the linear model QSPRMLR which was harshly evaluated by the statistical values: R2train = 0.864, Q2LOO = 0.840, SE = 0.711, Fstat = 131,355 and PRESS = 49.31. Furthermore, the artificial neural network model QSPRANN with architecture I(4)-HL(5)-O(1) was discovered with the same variables of the QSPRMLR model that the statistical results were extremely impressive as R2train = 0.970, Q2CV = 0.984 and Q2test = 0.974. Also, both of the QSPR models were externally validated on the data set of 18 logb11 values of independently experimental complexes. As a consequence, the results from the QSPR models could be used to calculate the stability constants of other new metal-thiosemicarbazones.