Levornidazole, a nitroimidazole compound, has been linked to hepatotoxic adverse effects in clinical settings. However, the hepatotoxicity of levornidazole and its impurities has not been fully elucidated. This study aimed to predict and evaluate the potential hepatotoxicity of levornidazole, and elucidate the underlying mechanisms of action. Computational models based on support vector machines (SVM) and artificial neural networks (ANN) predicted that levornidazole, ornidazole, and impurity II exhibited hepatotoxic effects. The hepatotoxicity of levornidazole and impurity II was confirmed using a zebrafish toxicity study, with impurity II demonstrating hepatotoxicity at lower doses. Molecular structure analysis revealed that the electronegativity of the side-chain groups and the molecular polarity structure were correlated with the degree of hepatotoxicity. The toxic response was primarily associated with specific structural domains of the molecule, including the 2-methyl-5-nitro-1H-imiddaster-1-yl structure and the substituent groups of 1-chloro and 2(S)-2-methyloxirane. Transcriptome sequencing analysis indicated that levornidazole and impurity II affect multiple metabolic processes in the liver, including glucose, lipid, protein, hormone, and drug metabolism. These findings highlight the potential hepatotoxic risks associated with levomeprazole and its impurities, emphasizing the importance of further investigation and regulatory attention to ensure patient safety.