Prediction of Log(IGC50)−1 for Benzene Derivatives to Ciliate Tetrahymena pyriformis from Their Molecular Descriptors

The purpose of this study was to develop the structure–toxicity relationships for a large group of organic compounds including 392 substituted benzenes to the ciliate Tetrahymena pyriformis (Log(IGC50)−1) using interpretable molecular descriptors. These descriptors were calculated using DRAGON and CODESSA software. Multiple linear regression and artificial neural network methods were used as linear and nonlinear feature-mapping techniques. The best obtained model was derived by MLR with seven descriptors which are: the molecular weight, the radial distribution function, the Kier shape index, the 26th component of atom-centered descriptors type of R–CX–R, the topographic electronic index, the H atoms attached to CO groups, the 24th component of atom-centered descriptors of R–CH–R. These descriptors can encode different features of molecules which are responsible for their steric, electronic, and lipophilicity interactions. The best obtained model had statistics of R2 = 0.822, F = 1386.806, and SE = 0.312 for training and R2 = 0.815, F = 361.384, and SE = 0.337 for prediction. The presented model shows better statistical parameters in comparison with a previous model. The reliability of the model was evaluated by using the leave-many-out cross-validation method (Q2 = 0.819 and SPRESS = 0.32) as well as by y-scrambling.