Abstract
The extensive use of industrial chemicals poses a serious threat to aquatic species such as the salmon species, which, when consumed, can affect human beings via their dietary intake. Salmon fish is a vital source of protein for maintaining human health. The present study aims to estimate the toxicity of diverse chemicals using in silico-based global model involving three different salmon species: Salmo salar, Oncorhynchus kisutch, and Oncorhynchus tshawytscha encompassing the toxicity endpoint median lethal concentration (LC(50)). Primarily, a quantitative structure-activity relationship (QSAR) model is developed using molecular descriptors. QSAR model descriptors are integrated with the similarity and error-based measures of read-across to develop the read-across structure-activity relationship (RASAR) model. Another emerging dimensionality reduction modeling algorithm, arithmetic residuals in K-groups analysis (ARKA) is employed to enhance the model's degree of freedom. Model quality was improved by hybrid model development which combined the feature matrix of the QSAR model with those of the RASAR and ARKA descriptors. Finally, to attain more trustworthy results and address the limitations of individual models, a partial least square (PLS)-based stacking model is developed using the predicted response values of QSAR, RASAR, ARKA, and hybrid models as descriptors. The stacking model outperforms the quality of the individual models which is evident from the determination coefficient R(2) (0.713), leave-one-out cross-validated correlation coefficient (Q(2) (LOO):0.697), predictive R(2) (Q(2) (F1) (: 0.797),) Q(2) (F2) (0.795) and lower value of root mean square error of prediction RMSEp (0.652). Additionally, classification modelling was performed with the feature matrix of the QSAR model by employing both linear and non-linear approaches. The developed stacking model can thus be used in environmental risk assessment aiding in toxicity data-gap filling and design of safe and green chemicals.