Acute Toxicity of Quinoline Derivatives as Antibacterial Drug Candidates: In Silico and In Vivo Studies

Abstract

In this work, a series of quantitative structure-toxicity relationship (QSTR) models was developed using the Online Chemical Database and Modelling environment (OCHEM). Two datasets consisting of 1204 and 1330 compounds were used. Three machine learning methods were employed for modelling: the transformer convolutional neural network, the transformer convolutional neural fingerprint, and the associative neural network. Seven QSTR models were developed, and their predictive performance was evaluated using a fivefold cross-validation procedure. This evaluation resulted in q² values between 0.74 and 0.75 for the regression models and a balanced accuracy between 81% and 83% for the classification models. The external test sets achieved balanced accuracy scores of 87% for classification and q² values between 0.72 and 0.73 for the regression models. The experimental results for the six compounds were consistent with those predicted by the QSTR models and differed slightly between the two compounds. The results of in vivo and in silico toxicity studies of quinoline derivatives indicated that all compounds have slight or moderate toxicity according to the classification by Passino-Smith and low toxicity according to the GHS categories. A comparative analysis of acute toxicity studies (LD₅₀) of quinoline derivatives on the hydrobiont Daphnia magna along with a comparative analysis of the predicted toxicity values of several drugs and quinolines allowed classifying the studied compounds as low-toxic and confirmed their prospects for further study and potential use in the medical field, particularly as effective antimicrobial agents.