Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Journal Club
  • Published:

Machine learning

Machine-made chemistry

Nature Reviews Chemistry volume 9, page 806 (2025)Cite this article

Subjects

In recent years, neural network–based machine learning potentials (MLPs) have emerged as a crucial bridge between first-principles (ab initio) calculations and large-scale atomistic simulations. However, conventional MLPs are typically trained on data sampled from molecular dynamics (MD) near equilibrium configurations. While this approach can be effective for specific reactions, such models lack generality and chemical reactivity. They frequently perform poorly when bonds break, surfaces reconstruct, or complex reaction networks evolve. In essence, traditional MLPs are tailored surrogates that struggle to function as universal computational engines.

The performance gap between REICO + EMLP and conventional MD-based MLPs is striking. MD-MLPs, derived from equilibrium configurations sampled by molecular dynamics, often fail to reproduce transition states, adsorption geometries, or reaction energetics, and may even produce non-physical outcomes. By contrast, EMLP consistently reproduces DFT results across diverse scenarios, including dimer potential energy curves, nanocluster dynamics, large-molecule adsorption on metal surfaces, and free-energy profiles of reactions in both gas and liquid phases. With an average energy error of ~0.1 eV and a transition-state search success rate of 100%, EMLP demonstrates that an element-driven approach to data generation and learning can endow MLPs with both universality and chemical fidelity.

This is a preview of subscription content, access via your institution

Access options

Buy this article

  • Purchase on SpringerLink
  • Instant access to the full article PDF.

USD 39.95

Prices may be subject to local taxes which are calculated during checkout

References

Original article

  • Yang, C., Wu, C., Xie, W., Xie, D. & Hu, P. General reactive element-based machine learning potentials for heterogeneous catalysis. Nat. Catal. 8, 891–904 (2025)

    Article  Google Scholar 

Related article

  • Bozal-Ginesta, C. et al. Developing machine learning for heterogeneous catalysis with experimental and computational data. Nat. Rev. Chem. 9, 601–616 (2025)

    Article  PubMed  Google Scholar 

Download references

Author information

Authors and Affiliations

  1. Department of Chemistry, North Carolina State University, Raleigh, NC, USA

    Emmanuel Adu Fosu

  2. Department of Chemistry, Yale University, New Haven, CT, USA

    Jindou Yang

  3. Energy Sciences Institute, Yale University, West Haven, CT, USA

    Jindou Yang

Authors
  1. Emmanuel Adu Fosu
    View author publications

    Search author on:PubMed Google Scholar

  2. Jindou Yang
    View author publications

    Search author on:PubMed Google Scholar

Corresponding authors

Correspondence to Emmanuel Adu Fosu or Jindou Yang.

Ethics declarations

Competing interests

The authors declare no competing interests.

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Fosu, E.A., Yang, J. Machine-made chemistry. Nat Rev Chem 9, 806 (2025). https://doi.org/10.1038/s41570-025-00775-7

Download citation

  • Published:

  • Version of record:

  • Issue date:

  • DOI: https://doi.org/10.1038/s41570-025-00775-7

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing