Researchers report in Nature a generative-AI method that reverse-engineers molecules by proposing structures from desired properties. A team from New York University and the University of Florida says their neural network suggests viable candidates about 10 times faster than prior methods while retaining comparable accuracy. The advance could speed up drug discovery, battery research and other materials development, though lab validation remains essential.
AI Reverse‑Engineers Molecules 10 Times Faster, Researchers Report in Nature
Researchers have developed a new generative-AI technique that can reverse-engineer molecules — the groups of atoms that form medicines, batteries and many other materials — according to a paper published in Nature.
Background: Historically, many therapies and materials were discovered by chance or trial-and-error (penicillin is the classic example). As computational models advanced, scientists began a more directed approach: define the desired molecular behaviors or "properties" first, then use algorithms to propose structures that could deliver those properties.
How the New Method Works
The team, led by researchers at New York University and the University of Florida, trained a neural network to generate candidate molecular structures from specified target properties. According to the study, the model proposes potentially viable molecular structures about 10 times faster than existing approaches while maintaining comparable accuracy.
Why This Matters
Faster, accurate generation of candidate molecules can shorten the design–test cycle in areas such as drug discovery, battery materials and industrial chemistry. By reducing the time needed to find promising candidates, the approach could let scientists run more experiments, iterate designs more quickly and accelerate real-world applications.
Important Caveats
Computational proposals still require laboratory validation. The AI speeds up the candidate-generation step but does not replace experimental testing or safety and efficacy studies. Readers should consult the full Nature paper for technical details, benchmarks, and limits of the method.
Institutions: New York University and the University of Florida (lead authors); paper published in Nature.
Help us improve.
