Machine learning prediction of acute toxicity with in vivo experiments on tetrazole derivatives

Abstract

This study investigates the application of machine learning techniques to predict the toxicity of tetrazole derivatives, aiding in the identification of environmental risks from chemical exposure. Utilizing LD₅₀ data sourced from the scientific literature and the ChemIDplus database, regression models were developed to forecast acute intraperitoneal toxicity in mice. A machine learning regression model for acute intraperitoneal toxicity in mice was constructed and validated on a test dataset, achieving high accuracy (R² = 0.76 and MSE below 10⁻⁴) and surpassing most of the comparable literature models. Molecular descriptors were computed via Mordred software to explore quantitative structure–activity relationships, and additionally, the model's robustness was demonstrated by measuring the acute toxicity of tetrazole derivatives synthesized through the azido-Ugi reaction.