Korean researchers have developed the first artificial intelligence model capable of understanding fundamental chemical principles and independently determining molecular stability. The breakthrough is expected to significantly accelerate the traditionally time-consuming and costly molecular design process.
According to scientific sources on May 16, a research team led by Professor Kim Woo-youn of the Chemistry Department at Korea Advanced Institute of Science and Technology (KAIST) recently developed the "Riemannian Diffusion Model (R-DM)," an AI model that predicts molecular structures by independently understanding the physical laws governing molecular stability.
The research team represented molecular structures as a map-like terrain of "hills" and "valleys" based on energy levels, designing the AI to navigate toward the lowest-energy valleys.
R-DM completes molecules by avoiding unstable structures and finding the most stable states across this three-dimensional energy landscape. The model searches for minimum energy points on curved surfaces based on Riemannian geometry, a mathematical theory dealing with curved rather than flat spaces. The research team explained, "This is the result of AI learning the fundamental chemical principle that 'matter prefers the lowest energy state.'"
Experimental results showed R-DM achieved "chemical accuracy" up to 20 times higher than existing AI models. Prediction errors were reduced to levels nearly indistinguishable from precision quantum mechanical calculations.
The technology is expected to find applications across various fields including new drug development, next-generation battery materials, and high-performance catalyst design. Both smartphone battery lifespan and drug development success depend on how stably the atoms comprising materials bond together. Finding the optimal atomic arrangement for the most stable molecule is the core step in molecular design—and this AI can now perform that task independently.
The research team stated the model is expected to serve as an "AI simulator" that will dramatically accelerate research and development speeds. The technology is particularly promising for environmental and safety applications, as it can rapidly predict chemical reaction pathways even in situations where experimentation is difficult, such as chemical accidents or hazardous substance dispersal.
Professor Kim Woo-youn said, "This is the first case where artificial intelligence has understood fundamental chemical principles and independently determined molecular stability," adding that "this technology could fundamentally transform how new materials are developed."
The research findings were published in the international journal Nature Computational Science on May 2.