Abstract
The evaluation of the safety of chemical substances requires the identification of a safe dose, which has no adverse effects on humans. This is obtained through animal studies, with exposure prolonged for months. Repeated-dose toxicity is a term in toxicology and pharmacology referring to the highest tested dose of a substance, so-called No Observed Adverse Effect Level (NOAEL). Experimental data on NOAEL taken from the literature and the OpenFoodTox database (total n = 848). To speed up the processing of the enormous number of substances we are exposed to, in silico models are an attractive solution. Monte Carlo technique, incorporating the Las Vegas algorithm, was applied to develop models for repeated-dose toxicity in rats. Optimal descriptors were calculated using correlation weights for attributes of the Simplified Molecular Input Line Entry System (SMILES). Computational experiments were conducted 5 times, with splits obtained using the Las Vegas algorithm. Good predictive potential was observed for these models, with an average determination coefficient on the validation set of 0.77 ± 0.04.