en.Wedoany.com Reported - A research team from Friedrich-Schiller-Universität Jena, Helmholtz-Zentrum Berlin für Materialien und Energie, the Helmholtz Institute for Polymers in Energy Applications Jena, and the Swiss software company Zakodium Sárl has developed an artificial intelligence system named SECS, which can recommend possible molecular structures from raw spectral measurement data and evaluate their plausibility. The system is available as an open-access resource, and the findings have been published in the professional journal Nature Communications.
Structural elucidation is a critical step in chemical research, but it poses significant challenges when dealing with novel or complex molecules. Dr. Kevin Jablonka from the University of Jena noted that commonly used methods such as nuclear magnetic resonance (NMR) spectroscopy, infrared spectroscopy, or mass spectrometry each provide limited structural clues, and the chemical puzzle formed by numerous measurement signals must be correctly assembled. For new molecules that have never been described before, impurities can generate their own signals or mask the actual substance's signals, making analysis particularly difficult. The advantage of the new system lies in its ability to handle the most common impurity issues found in routine proton NMR spectra.
Adrian Mirza, the first author of the paper, explained that the new SECS system combines two artificial intelligence approaches. The model first learns to convert spectra and molecular structures into a common mathematical representation, after which an evolutionary algorithm gradually optimizes candidate molecules by adding or removing atoms and chemical bonds, repeatedly checking whether the results better match the measured data. Ultimately, the system presents a ranked list of possible structures, accompanied by similarity scores based on the chemical context.
In a benchmark test using different spectroscopic methods, SECS ranked the correct molecular structure first in over 80% of cases. In a direct comparison with humans, the system performed on par with participating experts when solving 20 challenging NMR problems. Nevertheless, the research team emphasizes that SECS is not intended to replace human expertise but rather to provide a useful second opinion. If the recommended solution is plausible and scores highly, it helps strengthen confidence in the analytical result; conversely, it warrants closer scrutiny.
The application's source code, model data, and beta version are all publicly available. The current web version is primarily designed for direct analysis of raw one-dimensional proton NMR data, with support for more spectral types and more complex raw data to follow.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
