The intersection of chemistry and artificial intelligence (AI) is a fascinating area that attracts a lot of attention in both research and industry. We talked to people working in the field about the potential of AI to revolutionize chemical research, but also concerns, (current) limitations, and ethical implications for chemical applications. We also asked for ideas to try or experiment with, as well as useful articles and videos for beginners and advanced users.
Professor Florence Vermeire, Department of Chemical Engineering, KU Leuven, Belgium, works on machine learning in the field of pharmaceutical design and development with a main focus on predicting chemical properties and reaction kinetics.
What fascinates you about AI?
The broad horizon of new opportunities!
The breakthroughs made in AI over the past decade have inspired chemists to achieve the impossible. It is fascinating to experience the creative integration of chemistry and physics in machine learning algorithms. It helps us to tackle exceedingly complex problems with the smallest datasets. I believe we will continue to be surprised by the application of AI in chemistry, leading to novel solutions that were previously unimaginable.
Is there anything we should fear?
Generally, there are many ethical concerns in AI-related to inclusion, bias, and discrimination. Although there are no immediate ethical issues for the application of AI in chemistry, similar concerns persist. Mindfulness regarding our models’ capabilities and limitations is crucial, especially in automation and control of chemical processes where safety prevails.
A critical assessment of the capabilities of machine learning models is important. While these models bring many new opportunities, they also have their limitations, primarily related to the availability of training data. In fields like chemistry and material science data are often scarce, biased, and uncertain. Prior to training machine learning models, a thorough dataset analysis is imperative. Don’t expect miracles from AI and try to understand why it fails in some cases.
Do you have something for our readers to try out or experiment with?
If you aim to predict the chemical properties of molecules based on their structure, I strongly recommend utilizing the open-source software Chemprop. Today, it stands as the most popular public repository on GitHub in the field of chemistry.
The software uses message-passing neural networks to convert molecular graph-based representations to learned representations. Subsequently, a feedforward network learns the relationship to chemical properties. The community is actively integrating new features, expanding the software to have multiple molecules and reactions as input up to predicting uncertainty.
You can read all about these latest features in our most recent publication:
- Chemprop: A Machine Learning Package for Chemical Property Prediction,
Esther Heid, Kevin P. Greenman, Yunsie Chung, Shih-Cheng Li, David E. Graff, Florence H. Vermeire, Haoyang Wu, William H. Green, Charles J. McGill,
ChemRxiv December 2023.
There are plenty of examples and even a YouTube video to help you get started.
Can you recommend a good article/video/website for beginners and one you enjoyed recently?
Comprehending the fundamental workings of machine learning models is crucial before delving into applications. When I begun my machine learning adventures, I found value in Andrew Ng’s lectures on YouTube. In his comprehensive sessions, he covers a variety of topics spanning from linear regression and data splitting to exploring diverse machine learning methods, including neural networks.
Considering applications in chemistry, I can recommend the following papers:
- Scientific discovery in the age of artificial intelligence,
Hanchen Wang, Tianfan Fu, Yuanqi Du, Wenhao Gao, Kexin Huang, Ziming Liu, Payal Chandak, Shengchao Liu, Peter Van Katwyk, Andreea Deac, Anima Anandkumar, Karianne Bergen, Carla P. Gomes, Shirley Ho, Pushmeet Kohli, Joan Lasenby, Jure Leskovec, Tie-Yan Liu, Arjun Manrai, Debora Marks, Bharath Ramsundar, Le Song, Jimeng Sun, Jian Tang, Petar Veličković, Max Welling, Linfeng Zhang, Connor W. Coley, Yoshua Bengio, Marinka Zitnik,
Nature 2023, 620, 47–60.
- Machine learning for molecular and materials science,
Keith T. Butler, Daniel W. Davies, Hugh Cartwright, Olexandr Isayev, Aron Walsh,
Nature 2018, 559, 547–555.
For practical learning and coding exercises, I recommend the notebooks authored by Edgar Ivan Sanchez Medina, Antonio del Rio Chanona, and Caroline Ganzer. These resources have proven incredibly useful for our students who have just began their own machine learning projects.
Is there anything else you would like to share with readers of ChemistryViews?
It is a challenge to be bilingual in the languages of both chemistry and AI. The rapid evolution of AI and machine learning necessitates adapting our education swiftly to keep up with innovation. Embracing AI’s integration into education is crucial, yet maintaining a strong foundation in chemistry shouldn’t be overlooked. Keep your coding skills close and your chemistry textbooks even closer.
Thank you very much for the insights.
back to “Opinions on AI & Chemistry”