Assessment of the rat acute oral toxicity of quinoline-based pharmaceutical scaffold molecules using QSTR, q-RASTR and machine learning methods

Abstract

Quinoline is a common pharmaceutical scaffold molecule known for its wide range of biological and pharmacological activities, including antimalarial, antitumor, and antibacterial effects. With the continuous discovery of new bioactivities, there is a growing demand for the design and development of novel quinoline-based drugs. However, drug development is time-consuming and costly, and traditional toxicity testing methods such as animal experiments are resource-intensive. In the context of the 3Rs (Replacement, Reduction, Refinement) principle in animal research, quantitative structure-activity/toxicity relationship (QSAR/QSTR) modeling has become one of the most widely used methods for drug design and validation. This study collected acute oral toxicity data in rat for 33 quinoline derivatives and established a transferable, reproducible and interpretable QSTR model based on 2D molecular descriptors, following the OECD principles for model validation. Both internal and external validations were performed. The results demonstrated that the model possesses high goodness-of-fit, strong robustness, and excellent predictive power. Applicability domain (AD) analysis showed that the model has a broad range of applicability. Furthermore, the model's predictive performance was verified and enhanced using quantitative read-across structure-toxicity relationship (q-RASTR) and machine learning (ML) methods. Mechanistic interpretation provided detailed insights into the relationships between molecular descriptors and toxicity. Notably, for the first time, the model was applied for a true external dataset consisting of 1995 molecules lacking experimental values, thereby validating its extrapolation ability. Overall, the developed QSTR model exhibits good stability and predictive performance, offering theoretical support for the risk assessment and rational design of quinoline-based compounds.

Keywords: Acute oral toxicity; Machine learning; QSTR; Quinoline; q-RASTR.