New AI Tool Predicts Drug‑Enzyme Interactions

Scientists have created a fresh artificial‑intelligence system that helps predict how medicines will interact with human enzymes called cytochrome P450. These enzymes are key to breaking down foreign chemicals in the body. Existing tools often look only at the whole protein sequence, missing critical spots that decide whether a drug will bind or be processed.

The new framework, named CYPMol, mixes several data sources. It pulls in specific protein positions that are known to influence binding and reaction, adds overall protein patterns learned from large language models, and fuses them with pre‑trained chemical fingerprints of drugs. All this is handled in one deep‑learning network.

Tests show that CYPMol does better than earlier methods when identifying which drugs are processed by nine human P450 enzymes. Its accuracy scores reach 0.82 for substrates and 0.73 for inhibitors, numbers that surpass the best previous models. The system also pinpoints exact sites on enzymes where chemical changes happen, and it can do this for over five hundred P450 proteins from animals, plants, and microbes.

Researchers have made the code and data freely available online. Anyone can download the package or try the web interface to run predictions on their own molecules.

By combining detailed protein knowledge with advanced language‑based learning and chemical embeddings, CYPMol offers a powerful tool for both designing safer drugs and engineering enzymes with new capabilities.