Multimodal Machine Learning Models on Aqueous Direct Photodisappearance Rate Constants of Chemicals Exhibit...

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Toxicology

Background:

• Direct photodisappearance rate constant (kd) is crucial for assessing chemical persistence in aquatic environments.
• Existing in silico models for kd prediction often overlook environmental variables, limiting their applicability.
• Developing accurate predictive models is essential for managing chemical risks efficiently.

Purpose of the Study:

• To develop advanced in silico models for predicting the direct photodisappearance rate constant (kd) of chemicals.
• To incorporate both molecular structures and environmental conditions into predictive models for enhanced accuracy.
• To establish a robust method for defining the applicability domain of these predictive models.

Main Methods:

• Compiled a comprehensive dataset of 1281 kd values for 304 chemicals.
• Developed multimodal machine learning models integrating molecular descriptors and environmental factors (light, pH, temperature, dissolved oxygen, concentration).
• Proposed a novel applicability domain characterization using feature-response landscape analysis.

Main Results:

• Multimodal models demonstrated a significant improvement in predictive accuracy, with the coefficient of determination increasing from 0.369 to 0.756 on the validation set compared to unimodal models.
• The developed models effectively account for the influence of environmental conditions on photodisappearance kinetics.
• The proposed applicability domain method provides a reliable way to assess model reliability for specific chemical contexts.

Conclusions:

• Multimodal machine learning models offer a superior approach for predicting chemical photodisappearance rates by integrating molecular and environmental data.
• These models, combined with applicability domain assessment, provide a powerful tool for evaluating the environmental persistence of chemicals.
• The findings support informed chemical management strategies and environmental risk assessment.