Abstract

Predicting the bioconcentration of chemical compounds is essential for assessing environmental risks and potential toxicological impacts. The bioconcentration factor (BCF) is a key parameter used to quantify the extent of bioconcentration. It represents the ratio of the concentration of a substance in an organism to its concentration in the surrounding water at equilibrium. Herein, we developed a robust multitask deep learning model using a binary-tree strategy to classify compounds into three bioconcentration categories including non-bioaccumulative (nBC, with BCF value < 500L/kg), weak bioaccumulative (weak-BC, with 500L/kg ≤ BCF value < 5000L/kg), and strong bioaccumulative (strong-BC, BCF value ≥ 5000L/kg). The model exhibited excellent predictive performance, achieving over 90% accuracy and an area under the curve (AUC) value of 0.95 for each binary classification task. The final ternary classification model attained an overall accuracy of 91.11%, with particularly high accuracy for non-BC and strong-BC compounds.