Causal-Inference Machine Learning for Unveiling Hydroxyl Radical Reactivity with Antibiotics in Water Purification.

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Water Treatment

Background:

• Hydroxyl radical (HO·) reactivity with organic pollutants is vital for water purification via advanced oxidation processes.
• Existing machine learning (ML) models for predicting HO· reactivity often lack interpretability.
• Understanding molecular factors influencing HO· reactivity is key to optimizing water treatment.

Purpose of the Study:

• To develop an interpretable ML framework to identify molecular factors governing HO· reactivity with antibiotic contaminants.
• To establish a causal discovery framework for rational water purification design.

Main Methods:

• Employed DFT-derived descriptors (constitutional, quantum-chemical, Abraham) to characterize antibiotics.
• Utilized an attention-driven feature interaction method for optimal feature subset generation.
• Applied SHapley Additive exPlanations (SHAP) and causal interface ML for interpretability and causal inference.

Main Results:

• Identified key molecular properties like volume-regulated electronic migration ability (V_VIP) and halogen-mediated electronic attraction (#X_ME).
• Developed an optimized random forest model with high prediction accuracy (experimental relative errors < 6%).
• Established a causal discovery framework applicable even to small datasets.

Conclusions:

• The interpretable ML framework successfully unveils intrinsic molecular factors of HO· reactivity.
• The developed causal discovery approach enables more rational design of water purification strategies.
• This work provides a robust foundation for advancing AOPs in water treatment.