Premium
Comparison between genetic algorithm‐multiple linear regression and back‐propagation‐artificial neural network methods for predicting the LD 50 of organo (phosphate and thiophosphate) compounds
Author(s) -
Kianpour Mina,
Mohammadinasab Esmat,
Isfahani Tahereh Momeni
Publication year - 2020
Publication title -
journal of the chinese chemical society
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.329
H-Index - 45
eISSN - 2192-6549
pISSN - 0009-4536
DOI - 10.1002/jccs.201900514
Subject(s) - thiophosphate , linear regression , chemistry , correlation coefficient , artificial neural network , mean squared error , quantitative structure–activity relationship , test set , regression , biological system , regression analysis , algorithm , statistics , mathematics , artificial intelligence , stereochemistry , computer science , organic chemistry , biology
The DFT‐B3LYP functional method on the 6‐31G* basis set was employed to optimize and calculate molecular descriptors of 76 organo (phosphates and thiophosphate) derivatives. The molecular descriptors were used to establish the quantitative structure‐toxicity relationship (QSTR) for the acute oral toxicity of studied compounds by multiple linear regression (MLR) and artificial neural network (ANN) methods. The best results were obtained with an ANN model trained with the back‐propagation (BP‐ANN) algorithm. The prediction accuracy for the external test set was estimated by the root mean square (RMS) error, square correlation coefficient ( R 2 ), and absolute average deviation (AAD) which were equal to 0.0248797, 0.9091, and 15.12187, respectively. It was specified that 90.91% of external test set was correctly predicted and the present model proved to be superior to the MLR model. Accordingly, the model developed in this study can be used to predict the oral acute toxicity of organo (phosphates and thiophosphate) derivatives, particularly for those that have not been experienced as well as new compounds. The analysis of the statistical parameters of models published in the literature showed that our model was more efficient.