In silico genotoxicity and carcinogenicity prediction for food-relevant secondary plant metabolites

Abstract Humans are exposed to thousands of different secondary plant metabolites which may have beneficial health effects, but numerous compounds may also have toxic potential. In the present study we have examined the genotoxic and carcinogenic potential of 609 food-relevant phytochemicals by using computer models for toxicity prediction. We developed a scoring method and combined the results of different models to increase the predictive power. A combination of the VEGA models SARpy, KNN, ISS, and CAESAR, and of the LAZAR model “ Salmonella typhimurium ” for genotoxicity prediction performed better than the single models regarding specificity and accuracy. Statistical evaluation of the combined model for carcinogenicity prediction was not possible due to the low number of substances suitable for model validation. The in silico results of the present exercise will be useful for priority setting purposes regarding future risk assessment of secondary plant metabolites. Based on our analysis, (-)-asimilobine, aloin, annoretine, chrysothrone, coptisine, elymoclavine, and thalicminine were predicted to be genotoxic with high probability and may therefore be selected for subsequent experimental genotoxicity testing. Moreover, the class of pyrrolizidine alkaloids is suggested to be a high priority subject for further studies as these substances have been predicted to be carcinogenic with high probability.