QSPR-based calculation model for stability constants of new metal-thiosemicarbazone complexes with hybrid techniques

In the present study, we calculated the logb12 stability constant of twenty new ML2 complexes between thiosemicarbazone and metal ions based on the modelling techniques of the quantitative structure and property relationship (QSPR). The QSPR models were developed by combining the genetic algorithm (GA) with multivariate linear regression techniques (QSPRGA-MLR), support vector regression (QSPRGA-SVR), and artificial neural network (QSPRGA-ANN). The descriptive parameters were calculated from semi-empirical quantum computation with the new versions PM7 and PM7/sparkle. The resulting QSPRGA-MLR models had three variables, and the QSPRGA-SVR and QSPRGA-ANN models were developed from the variables of the QSPRGA-MLR model. The results show that the best QSPRGA-SVR model had the following optimal parameters: C = 10.0; g = 0.333; e = 0.10 with 51 support vectors, and a QSPRGA-ANN model with the network architecture I(3)-HL(10)-O(1) was successfully developed. Furthermore, the quality of QSPR models conformed to statistical values ​​according to OECD principles and Tropsha’s criteria.