QSAR Studies of New Compounds Based on Thiazole Derivatives as PIN1 Inhibitors via statistical methods
Kamal Tabti · RHAZES: Green and Applied Chemistry · 2020
In this study, mathematical and statistical approaches to QSAR-2D modeling were used to predict biological activity against PIN1 and to explain the origin of the activity of these studied compounds, to design new thiazole derivatives with high predicted values against PIN1. In this regard, a series of thiazole derivatives of 25 compounds were analyzed by principal component analysis, linear regression, partial least squares analysis, and artificial neural network. The descriptors studied were selected from a set of descriptors (topological, electronic geometric and physicochemical), having a chemical explanatory meaning of molecular bioactive. The model predictive was validated by different methods of internal validation, external validation, and randomized Y-test. Moreover, the statistical indicators R², MSE were used to evaluate the predicted responses of the models compared to the observed data. A leveraged approach was used using the Williams plot to detect outliers and verify that such a chemical compound is included in the applicability of the models developed or not. Thus, the model developed by the MLR method showed satisfactory performance during the learning and validation and test phase with four descriptors: MR, LogP, ELUMO, and J; the exception of the PLS method which did not pass certain external validation criteria. Therefore, the best model is RLM with R² = 0.76, MSE = 0.039, a cross-validation coefficient (R²cv = 0.63) and an external predictive power (R²test = 0.78). The result of the ANN model with the Levenberg-Marquart algorithms showed us better performance with the architecture [4-10-1]: R² = 0.98, R²cv = 0.99, R²test = 0.98, MSE = 0.013.