For more than a decade, MIT Associate Professor Rafael Gomez-Bombarelli has used artificial intelligence to create new materials. As technology has expanded, so have their ambitions.
Now, the newly appointed professor in materials science and engineering believes AI is set to transform science in ways that were never possible before. His work at MIT and beyond is dedicated to accelerating that future.
“We are at another turning point,” says Gomez-Bombarelli. “The first was around 2015 with the first wave of representation learning, generative AI, and high-throughput data in some areas of science. These are some of the technologies I first brought to my lab at MIT. Now I think we’re at the second inflection point, merging language and multiple modalities into general scientific intelligence. We have all the model classes we need to reason about language, to reason over physical structures, and to reason over synthesis recipes.” And there will be scaling laws.”
Gómez Bombarelli’s research combines physics-based simulations with approaches such as machine learning and generative AI to discover new materials with real-world applications. His work has led to new materials for batteries, catalysts, plastics and organic light-emitting diodes (OLEDs). He has also co-founded several companies and serves on the scientific advisory board for startups applying AI in drug discovery, robotics, etc. His latest company, Leela Sciences, is working to build a scientific observability platform for the life sciences, chemical and materials sciences industries.
All that work is designed to ensure that the future of scientific research is more seamless and productive than today’s research.
“AI for science is one of the most exciting and ambitious uses of AI,” says Gomez-Bombarelli. “Other applications for AI have more downsides and ambiguity. AI for science is about bringing about a better future over time.”
From experiments to simulations
Gómez-Bombarelli grew up in Spain and was attracted to physics from an early age. In 2001, he won the Chemistry Olympic Competition, setting him on an academic track in chemistry, which he studied as an undergraduate at his hometown college, the University of Salamanca. Gómez-Bombarelli stayed on for his PhD, where he investigated the action of chemicals that damage DNA.
“My PhD started experimentally and then halfway through I got bit by the simulation and computer science bug,” he says. “I started simulating the same chemical reactions I was measuring in the lab. I love the way programming organizes your brain; it seems like a natural way to organize one’s thinking. Programming is also much less limited in what you can do with your hands or with scientific instruments.”
Next, Gomez-Bombarelli went to Scotland for a postdoctoral position, where he studied quantum effects in biology. Through that work, he connected with Alan Aspuru-Guzik, a chemistry professor at Harvard University, whom he joined in 2014 for his next postdoc.
“I was one of the first people to use generative AI for chemistry in 2016, and I was on the first team to use neural networks to understand molecules in 2015,” says Gomez-Bombarelli. “These were the early, early days of deep science education.”
Gómez-Bombarelli also began working to eliminate the manual parts of molecular simulations in order to run more high-throughput experiments. He and his colleagues performed hundreds of thousands of calculations across a variety of materials and discovered hundreds of promising materials to test.
After two years in the laboratory, Gómez-Bombarelli and Aspuru-Guzik started a general purpose materials calculation company, which eventually focused on the production of organic light-emitting diodes. Gomez-Bombarelli joined the company full-time and describes it as the hardest thing he has ever done in his career.
“It was amazing to make something tangible,” he says. “Also, after watching Aspuru-Guzik run a lab, I didn’t want to be a professor. My dad was a professor in linguistics, and I thought it was a cool job. Then I saw Aspuru-Guzik with a group of 40 people, and he was living on the road 120 days a year. It was crazy. I didn’t think I had that kind of energy and creativity. Is.”
In 2018, Aspuru-Guzik suggested Gomez-Bombarelli apply for a new position in MIT’s Department of Materials Science and Engineering. But, due to his nervousness about the faculty job, Gomez-Bombarelli let the deadline pass. Aspuru-Guzik confronted him in his office, slammed his hands on the table and told him, “You have to apply for this.” This was enough for Gomez-Bombarelli to submit a formal application.
Luckily at his startup, Gomez-Bombarelli had spent a lot of time thinking about how to create value from computational content discovery. During the interview process, he says, he was attracted by the energy and collaborative spirit at MIT. He also began to appreciate the possibilities of research.
“What I was doing as a postdoc and at the company was a subset of what I could do at MIT,” he says. “I was making products, and I still got to do that. Suddenly, my universe of work became a subset of this new universe of things I could discover and do.”
It has been nine years since Gomez Bombarelli attended MIT. Today his laboratory focuses on how the structure, composition, and reactivity of atoms affect the performance of materials. He has also used high-throughput simulations to create new materials and helped develop tools to merge deep learning with physics-based modeling.
“Physics-based simulations create data, and AI algorithms get better the more data you give them,” says Gomez Bombarelli. “There are all kinds of virtuous cycles between AI and simulation.”
The research group he created is entirely computational – they don’t run physical experiments.
“It’s a blessing because we can have a lot of expansion and we can do a lot of things at once,” he says. “We love working with experimentalists and try to be good partners with them. We also love creating computational tools that help experimentalists test ideas coming from AI.”
Gomez-Bombarelli is still focused on real-world applications of the materials he invented. His laboratory works closely with companies and organizations such as MIT’s Industrial Liaison Program to understand the material needs of the private sector and the practical barriers to commercial growth.
accelerating science
As excitement around artificial intelligence has grown, Gomez-Bombarelli has seen the field mature. Companies like Meta, Microsoft, and Google’s DeepMind now regularly conduct physics-based simulations, which is reminiscent of what he was working on in 2016. In November, the US Department of Energy launched the Genesis mission to accelerate scientific discovery, national security and energy dominance using AI.
“AI for simulation has changed from something that could maybe work to a consensus scientific approach,” says Gomez-Bombarelli. “We’re at an inflection point. Humans think in natural language, we write papers in natural language, and it turns out that these big language models that master natural language have opened up the ability to speed up science. We’ve seen that scaling works for simulation. We’ve seen that scaling works for language. Now we’re going to see how scaling works for science.”
When he first came to MIT, Gomez-Bombarelli says he was stunned to see how non-competitive things were between researchers. He tries to bring that same positive mindset to his research group, which is made up of about 25 graduate students and postdocs.
“We have naturally evolved into a very diverse group with different types of mindsets,” says Gomez-Bombarelli. “Everyone has their own career aspirations and strengths and weaknesses. Figuring out how to help people become the best version of themselves is fun. Now I’m pushing for people to apply for faculty positions after the deadline. I feel like I’ve passed that baton on.”
