QSAR-QSIIR-based prediction of bioconcentration factor using machine learning and preliminary application.

Area of Science:

• Environmental Chemistry
• Toxicology
• Computational Chemistry
• Machine Learning

Background:

• Bioconcentration factor (BCF) is crucial for human health ambient water quality criteria (HHAWQC).
• Experimental BCF determination is costly and time-consuming.
• Existing Quantitative Structure-Activity Relationship (QSAR) models have limitations in scope and accuracy for diverse pollutants.

Purpose of the Study:

• To develop a robust QSAR-QIIR model for predicting BCF across multiple chemical substances and aquatic species.
• To improve the accuracy and applicability of BCF prediction models.
• To derive HHAWQC for BTEX in China using the developed model.

Main Methods:

• Selection of 17 molecular descriptors and 5 bioactivity descriptors from extensive datasets.
• Construction of a QSAR-QIIR model using an optimized 4-MLP machine learning algorithm.
• Validation of the model using verification and test sets, achieving high R2 values (0.8575 and 0.7924).

Main Results:

• The developed QSAR-QIIR model demonstrates significantly improved prediction accuracy for BCF.
• Predicted BCF values closely match measured values, with differences mostly within 1.5 times.
• BCF for BTEX in Chinese aquatic products was successfully predicted.

Conclusions:

• The novel QSAR-QIIR model offers a reliable and efficient method for BCF prediction for diverse chemicals and species.
• The model facilitates the derivation of HHAWQC, providing a valuable reference for water quality standards.
• This approach supports environmental risk assessment and regulatory development for chemical pollutants.