AI Predicts Biochar Efficiency: Water Purification from Antibiotics Accelerates Significantly

Scientists have created an AI model that accurately predicts how quickly biochar decomposes antibiotics in water — this reduces filter development from years to weeks and paves the way for affordable drinking water purification.

Researchers led by Junaid Latif and Na Chen collected data from dozens of previous studies, identifying 16 key parameters: the type of raw material for biochar, pyrolysis temperature, porosity, chemical composition, concentration of oxidants, and reaction conditions.

Based on this data, they trained several machine models, and the TabPFN transformer architecture showed the best accuracy: a coefficient of determination R² ≈ 0.91 with minimum prediction error.

The scientific novelty lies in the transition from a trial-and-error method to the targeted design of materials. Previously, selecting effective biochar required hundreds of laboratory experiments; now the algorithm evaluates in seconds which combination of parameters will yield the maximum pollutant degradation rate.

In parallel, the team created a web tool: any researcher can enter their material's parameters and instantly receive a prediction of reaction kinetics.

The practical significance is obvious: antibiotics in wastewater are one of the drivers behind the growth of bacterial resistance to drugs.

Biochar derived from agricultural waste (corn cobs, straw) is cheap and eco-friendly, and when combined with photocatalysts (TiO₂, g-C₃N₄), it is capable of decomposing sulfonamide antibiotics by 98% in 60 minutes under the influence of sunlight.

The AI model helps find the exact formulation that will ensure such a result in specific water treatment conditions.

It is important to understand the limitations. So far, the model has been trained on data from a limited set of antibiotics and laboratory conditions; real wastewater contains a mixture of pollutants, organic matter, and suspended solids that can affect the process.

Also, the long-term stability of composites on an industrial scale requires additional verification — in laboratory tests, the material maintained its efficiency after five cycles, but industrial operation may be harsher.

What's next? The team plans to expand the dataset to include data on other classes of pollutants — PFAS, microplastics, and next-generation pharmaceuticals.

In parallel, negotiations are underway with pilot water treatment plants to test AI-designed biochar filters in field conditions. If the results are confirmed, the technology could become part of standard solutions for municipalities and industrial enterprises within the next 3–5 years.