DBCata AI Model Speeds Up Catalyst Screening
Researchers have built a new AI model that could speed up the search for better catalysts. The model is called DBCata and it was described in Nature Communications on November 21. Catalysts speed up chemical reactions, but finding the best material usually needs many expensive quantum calculations. DBCata learns to predict near-equilibrium adsorption structures directly, without needing explicit energy or force labels. It combines a periodic Brownian-bridge framework with an equivariant graph neural network to map unrelaxed structures to relaxed ones. In tests, the model helped screen alloy catalysts for the oxygen reduction reaction more quickly. That reaction matters for fuel cells and clean energy devices. The authors say the approach could cut the time and cost of high-throughput screening across chemistry.