Risk assessment of aromatic organic chemicals to T. pyriformis in environmental protection using regression-based QSTR and Read-Across algorithm

Hazardous aromatic compounds have a high probability of entering the surrounding environment, posing a threat to humans and other habitats. It is, thus, necessary to estimate the toxicity of these chemicals as a pre-ventive measure before being marketed. The present study involves the development of in silico-based predictive 2D-QSTR models using a PLS regression approach for the exploration of the structural features responsible for the toxicity of T. pyriformis using simple and easily interpretable 2D descriptors employing a dataset containing 892 chemicals. The developed model was extensively validated by evaluating the model's reliability and predict-ability using internationally accepted external and internal validation metrics. We have also used the Intelligent Consensus Predictor tool and Read-Across software, which shows better results for test set compounds as compared with individual PLS models. We have also validated the models using a set of 383 external set com-pounds which are not used for the development of models. The results suggested that molecules with aliphatic aldehyde groups are much more toxic to the protozoan whereas chemicals containing C-N fragments at the topological distance 3 & 4, polar and alcoholic groups are less toxic to the protozoan. In conclusion, the developed QSTR and Read-Across models might be extremely beneficial as guides for researchers to estimate the toxicity profile of novel compounds against T. pyriformis.

Automated summary

Hazardous aromatic compounds are likely to enter the surrounding environment and pose a threat to humans and other habitats. Therefore, it is necessary to assess their toxicity before being marketed. This study developed in silico-based predictive models using a PLS regression approach to explore the structural features responsible for the toxicity of T. pyriformis. The models were validated using external and internal validation metrics, and Read-Across software showed better results for test set compounds compared to individual models.