Assessment of fine-tuned large language models for real-world chemistry and material science applications
2025

Using Large Language Models in Chemistry

Sample size: 22 publication Evidence: high

Author Information

Author(s): Van Herck Joren, Gil María Victoria, Jablonka Kevin Maik, Abrudan Alex, Anker Andy S., Asgari Mehrdad, Blaiszik Ben, Buffo Antonio, Choudhury Leander, Corminboeuf Clemence, Daglar Hilal, Elahi Amir Mohammad, Foster Ian T., Garcia Susana, Garvin Matthew, Godin Guillaume, Good Lydia L., Gu Jianan, Xiao Hu Noémie, Jin Xin, Junkers Tanja, Keskin Seda, Knowles Tuomas P. J., Laplaza Ruben, Lessona Michele, Majumdar Sauradeep, Mashhadimoslem Hossein, McIntosh Ruaraidh D., Moosavi Seyed Mohamad, Mouriño Beatriz, Nerli Francesca, Pevida Covadonga, Poudineh Neda, Rajabi-Kochi Mahyar, Saar Kadi L., Hooriabad Saboor Fahimeh, Sagharichiha Morteza, Schmidt K. J., Shi Jiale, Simone Elena, Svatunek Dennis, Taddei Marco, Tetko Igor, Tolnai Domonkos, Vahdatifar Sahar, Whitmer Jonathan, Wieland D. C. Florian, Willumeit-Römer Regine, Züttel Andreas, Smit Berend

Primary Institution: École Polytechnique Fédérale de Lausanne (EPFL)

Hypothesis

Can fine-tuned large language models (LLMs) effectively predict chemical properties and outcomes?

Conclusion

Fine-tuning LLMs can lead to predictive models that outperform traditional machine learning methods in various chemical applications.

Supporting Evidence

  • Fine-tuned LLMs showed superior performance in predicting chemical properties compared to traditional models.
  • LLMs can effectively handle unstructured data, making them versatile for various chemical applications.
  • Fine-tuning with small datasets can still yield predictive models.
  • LLMs can simplify the process of predicting chemical outcomes without extensive feature engineering.
  • Natural language input allows for broader accessibility in chemical research.
  • LLMs demonstrated high accuracy in multiple case studies across different chemical domains.
  • Fine-tuning LLMs can significantly reduce the need for computational resources in experimental setups.
  • LLMs can learn from existing literature, providing insights into chemical systems.

Takeaway

This study shows that computers can learn to answer chemistry questions by reading lots of text, helping scientists make better predictions without needing to do as many experiments.

Methodology

The study involved fine-tuning three open-source LLMs on 22 datasets related to various chemical questions and benchmarking their performance against traditional machine learning models.

Potential Biases

Potential bias due to the reliance on existing datasets which may not represent all chemical scenarios.

Limitations

The performance of LLMs may not be optimal for every study, especially with unbalanced datasets.

Digital Object Identifier (DOI)

10.1039/d4sc04401k

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication