Simulation of fish chronic toxicity using the Las Vegas algorithm and the vector of ideality of correlation

The presence of active pharmaceutical ingredients in aquatic environments calls for greater attention to the risks that these substances may pose, particularly concerning chronic toxicity. The no observed effect concentration is a commonly used endpoint to summarize chronic toxicity in ecological risk assessment and ecotoxicology. A new approach is proposed here for predicting the no observed effect concentration in fish according to Test Guideline 210 of the Organisation for Economic Co-operation and Development ("Early-Life Stage Toxicity Test"; tested on >200 substances) through use of a molecular questionnaire. Molecular questioning is a survey of molecules represented as Simplified Molecular Input Line Entry System (SMILES) for the presence of individual features extracted from SMILES. The Las Vegas algorithm was used to rationally split the data into training and validation sets. The optimal descriptors for one-parameter linear regression were calculated by the Monte Carlo method based on the index of ideality of correlation. The statistical quality of the model is as follows: the coefficient of determination is about 0.7 for the training set and 0.7 for the validation set of the substances studied. In addition, we used several machine learning algorithms, but the results were poor. The model presented here compares favorably with models previously published for the same endpoint and chemicals. Since active pharmaceutical ingredients are complex molecules that can have multiple effects, particularly on human health, these results are of particular interest.