Generative AI (GenAI) has the potential to reshape science and enable major breakthroughs across a wide range of disciplines. I have explored this vision in depth in What Generative AI can(not) do for Computational Science, Generative AI for Computational Science: A Viable Path, and Generative AI for Computational Science by Example. These posts provide the broader context for my thinking on the future direction of GenAI and the application areas where I believe it can create the most value.
One field where GenAI has already enabled major breakthroughs is computational chemistry. Computational chemistry is one of the fields that holds the greatest promise for the future and prosperity of humanity. Perhaps we should define it first. Computational chemistry involves the use of numerical simulations, algorithms, and rigorous chemistry to predict properties of atoms, molecules, large biomolecules, and medicinal compounds. Examples of classical computational chemistry tasks include molecular structure prediction, computational drug design, the design of new materials, and even quantum-mechanical orbital calculations. More broadly, computational chemistry is a class of methods and techniques used for in silico design of drugs, compounds, materials and proteins. ‘In Silico’ indicates ‘by means of computational tools’ and is a third class of experimentation besides in vivo (in the live organism) and in vitro (in a lab).
Perhaps when people think of (generative) AI in chemistry, AlphaFold is usually the first breakthrough that comes up, and for good reason. AlphaFold showed that artificial intelligence can solve large-scale molecular prediction problems at a level that genuinely moved the needle in science. It marked a historical breakthrough in protein structure prediction, and redefined what is possible by predicting 3D protein structures with high accuracy and speed - a task that traditional computational methods are not up to (yet).
The breakthroughs enabled by Generative AI don’t stop at protein folding. Building on AlphaFold, RFdiffusion extends from prediction to design, making it possible to generate entirely new proteins and molecular structures with targeted properties and functions.
These successes are not isolated achievements: GenAI is starting to become a central technology for molecular discovery, drug design, and engineering. Even more, many research articles have come out over the last three years exploring all kinds of generative AI models for chemistry.
The Next Stage of my Blog
Computational chemistry was, in many ways, my first love in computational research. Long before generative models and scientific machine learning became a thing, I always wanted to apply computational science to chemistry. There is something uniquely compelling and productive about the field: applying numerical algorithms to design molecules, drugs and material with certain chemical properties to make better products, to cure untreatable diseases and to advance humanity. Computational chemistry was what I always wanted to do with my life, and I finally see the opportunity to go back to it.
For that reason, I want to devote my time to the intersection of Generative AI and computational chemistry. In particular, I want to pursue three goals.
- Dive deeply into the literature on computational chemistry and GenAI, with the concrete goal of reading 100 papers in the field by the end of the year, while also exploring the software, tools, and products that already exist;
- Develop and test my own research ideas in generative computational chemistry, with the longer-term ambition of turning them into a real software product and a company;
- Share the entire journey on Substack, including my take on the literature and current state of the field.
In other words, I want to combine reading, writing, and building. I want to understand which approaches are genuinely useful, where the field still falls short, and which open problems might lead to real breakthroughs. Most importantly, I want this project to be shared. So if this topic excites you too, I would be delighted to have you along for the journey.
I hope to meet all of you along the journey!
P.S. My explorations of physics-informed learning are not yet over, more about that journey later this week.