APPLICATION OF QSPR APPROACH FOR DEVELOPMENT OF NOVEL METAL-THIOSEMICARBAZONE COMPLEXES

Abstract

Twenty novel metal-thiosemicarbazone complexes (ML2) were calculated the stability constants (log12) based on the quantitative structure-property relationship (QSPR) models. The QSPR models were developed using multivariate linear regression (MLR), support vector regression (SVR), and artificial neural network (ANN) methods. Descriptors of the models were calculated from the PM7 and PM7/sparkle semi-empirical quantum mechanisms. The quality of the QSPR models was tightly controlled by the statistical values of OECD instructions and Tropsha’s standards. As a result, the best QSPRMLR model includes five variables: Dipole, xv2, xch5, SHBa, and 5C, with statistical values such as R2train = 0.922, Q2LOO = 0.861, and RMSE = 0.759. Besides, the best QSPRSVR model consists of capacity C = 10.0, gamma  = 0.10, and epsilon  = 0.1 with the number of support vectors equal to 42 and suitable regression parameters: R2 = 0.925, and RMSECV = 0.536. The QSPRANN model with network architecture I(5)-HL(6)-O(1) and exponential transfer function was trained from descriptors of the MLR model and showed impressive results as R2train = 0.986; Q2test = 0.876 and Q2validation = 0.921. In addition, this study used an external validation (EV) dataset of 25 log12 experimental values to build complete QSPR models with Q2EV-MLR, Q2EV-SVR, and Q2EVANN values of 0.834, 0.865, and 0.881, respectively. The positive results of the models can be used to find other new thiosemicarbazone and their complexes for applications in chemical, analytical, and environmental fields.