Modeling the toxicity of amide herbicides using the electrotopological state

The toxicity (rat, acute, oral LD50) of a set of 50 amides that serve as herbicides is modeled using the electrotopological state indices (E‐state) with multiple linear regression. A five‐variable model produces sound statistics: r2 = 0.75, s = 0.23. The cross‐validation statistics indicate a model useful for prediction: r2press = 0.65, spress = 0.27. Further, a set of nine compounds was set aside for testing the model. The mean absolute error for the test set is MAE = 0.27, supporting the claim that the model may be used for prediction. Structure interpretation is given for each variable so that the model may be used to assist in design of molecules with lower toxicity.