New AI method captures long-range atomic interactions in complex molecules

Researchers from Google DeepMind in Berlin, BIFOLD, and the Technical University of Berlin have introduced a new machine learning method, Euclidean Fast Attention, that enables global atomic interactions in chemical systems to be. The institutional report frames the development in practical terms and ties it to the broader mission or observing effort.

The significance lies in chemistry gains force when a claimed structure or process can be described with enough precision to be reproduced by others. Synthetic routes, spectroscopic signatures, yield under defined conditions and stability under realistic operating parameters are the currency of credibility in chemistry, and a result that lacks these details cannot be evaluated independently. The distance between a discovery on a laboratory bench and a process that works reliably at scale is measured in years of optimization, and each step reveals constraints that were invisible at smaller scale. Researchers from Google DeepMind in Berlin, BIFOLD, and the Technical University of Berlin have introduced a new machine learning method, Euclidean Fast Attention (EFA)—that. By Jean-Paul Olivier, Berlin Institute for the Foundations of Learning and Data, BIFOLD This article has been reviewed according to Science X's editorial process and policies.

Editors have highlighted the following attributes while ensuring the content's credibility: Add as preferred source New AI sees the entire molecule: Unlike previous methods, which. BIFOLD Researchers from Google DeepMind in Berlin, BIFOLD, and the Technical University of Berlin have introduced a new machine learning method, Euclidean Fast Attention.

This could allow chemical and materials science processes to be simulated more accurately in the future, potentially accelerating the development of new drugs, more efficient. The work, titled "Machine learning global atomic representations with Euclidean fast attention," was published in Nature Machine Intelligence in March 2026.

Such simulations form the foundation of modern drug development, as well as the design of new materials and more efficient catalysts. This results in a highly complex many-body system in which even small changes in one location can affect the behavior of the entire system.

The broader interest lies in whether the claimed property or reaction pathway can be characterized with enough precision to support replication by other groups. Chemistry has a replication problem that is less discussed than the one in psychology or medicine, but it is real: synthetic procedures that work reliably in one laboratory sometimes fail to transfer, for reasons ranging from impure starting materials to undocumented temperature sensitivities. A result that comes with full experimental detail and a clear characterization of the product is far more valuable than one that reports a discovery without the procedural backbone.

This is exactly where the research team's new method comes into play. In their experiments, the researchers show that EFA effectively captures different long-range effects and can describe chemical interactions for which conventional.

Because the account originates with Phys. org Chemistry, it functions best as a primary institutional report that is close to the data and operations, not as independent scientific validation. Institutional communications are produced by organizations with legitimate interests in presenting their work in a favorable light, which does not make them unreliable but does make them partial. Details that complicate the narrative, including instrument limitations, unexpected failures and results below projections, tend to be minimized relative to progress messages. Technical documentation and peer-reviewed publications, where they exist, provide the complementary layer that institutional releases cannot substitute.

The next step is to see whether independent groups working with orthogonal techniques reach compatible conclusions, and whether the result scales beyond the conditions used in the original study. Chemical discoveries that matter tend to be ones whose key properties can be measured by multiple spectroscopic, crystallographic or computational methods that are unlikely to share the same blind spots. Scalability, cost and long-term stability under realistic operating conditions are additional filters that come into play before any practical application becomes viable.

Source