A new artificial‑intelligence system can instantly evaluate the toxicity of hundreds of chemical by‑products formed during water disinfection. Trained on a curated dataset of known compounds, the model predicts hazardous effects for over a thousand disinfection by‑products, giving utilities and researchers a rapid, cost‑effective tool to prioritize safety testing and protect public health.
The Challenge of Disinfection By‑products
Municipal water treatment relies on chlorine‑based disinfectants to destroy pathogens. When chlorine reacts with natural organic matter such as fulvic acid, it creates a complex mixture of disinfection by‑products (DBPs) including trihalomethanes, haloacetic acids, and many lesser‑known halogenated organics. Numerous DBPs are classified as carcinogenic or otherwise harmful, prompting strict regulatory limits on their concentrations in drinking water.
Sources and Health Risks
Traditional toxicity assessment of DBPs depends on laboratory bioassays that are labor‑intensive, costly, and time‑consuming. This bottleneck limits the ability of water utilities and scientists to keep pace with the expanding catalog of DBPs generated by evolving treatment practices.
How the AI Model Predicts Toxicity
Deep‑Learning Architecture
The model was trained on a curated library of DBP structures paired with experimentally measured toxicological endpoints. By learning the relationships between molecular features and biological effects, the deep‑learning network can infer the toxicity of untested compounds with high confidence.
High‑Throughput Screening Benefits
In its initial run, the system generated toxicity predictions for 1,163 DBPs that are either already detected in treated water or are plausible by‑products of common disinfection processes. The AI does not replace laboratory testing; instead, it serves as a rapid screening tool to prioritize which compounds merit detailed experimental scrutiny.
Complementary Advances in Water Treatment
Photocatalytic Degradation of Precursors
Emerging photocatalysts based on bismuth oxychloride combined with two‑dimensional MXene materials accelerate the breakdown of fulvic acid under visible light, removing the organic precursors that give rise to harmful DBPs. These catalysts retain high activity over multiple reuse cycles, offering a practical approach to reduce DBP formation at the source.
Impact on Regulators, Utilities, and Public Health
Faster Risk Assessment
By flagging high‑toxicity candidates early, water utilities can adjust treatment parameters—such as chlorine dosage, contact time, or alternative disinfectants—to minimize the formation of the most hazardous DBPs. Regulators can leverage the model’s predictions to refine safety thresholds and focus monitoring efforts on compounds that have previously escaped routine testing.
Open‑Access Collaboration
The model is offered as an open‑access platform, encouraging researchers to contribute new experimental data and iteratively improve predictive accuracy. This collaborative framework accelerates the discovery of hidden chemical threats and supports evidence‑based policy development.
Future Outlook
Validation and Integration
While the AI tool provides speed and scale, its outputs must be validated through conventional toxicology studies. Ongoing integration of AI‑driven screening with advanced oxidation technologies promises a data‑centric, proactive approach to water safety, helping utilities address aging infrastructure, source variability, and tightening health standards.
