For more than a decade, MIT Associate Professor Rafael Gómez-Bombarelli has been using artificial intelligence to create new materials. As technology develops, so do his ambitions.
Now, a newly appointed professor of materials science and engineering believes that artificial intelligence has the ability to transform science in ways that have never been possible before. His work at MIT and beyond is dedicated to accelerating that future.
“We are at the second turning point” – says Gómez-Bombarelli. “The first one was around 2015, with the first wave of representation learning, generative AI, and high-throughput data in certain areas of science. These are some of the techniques that I first introduced to my lab at MIT. Now I think we're at a second inflection point, mixing language and combining multiple modalities into a general scientific intelligence. We'll have all the classes of models and scaling laws needed to infer language, infer material structures, and infer synthesis recipes.
Gómez Bombarelli's research combines physics-based simulations with approaches such as machine learning and generative artificial intelligence to discover new materials with promising real-world applications. His work has resulted in new materials for batteries, catalytic converters, plastics, and organic light-emitting diodes (OLEDs). He has also co-founded numerous companies and served on scientific advisory boards for startups applying artificial intelligence to drug discovery, robotics, and more. His newest company, Lila Sciences, is working to build a scientific superintelligence platform for the life sciences, chemical and materials sciences industries.
All of this work is intended to ensure that future research will be more seamless and productive than current research.
“AI in science is one of the most exciting and ambitious applications of AI,” says Gómez-Bombarelli. “Other applications of artificial intelligence have more flaws and ambiguities. Artificial intelligence in science is about ensuring a better future in a timely manner.”
From experiments to simulations
Gómez-Bombarelli grew up in Spain and was interested in physical sciences from an early age. In 2001, he won the competition at the Chemistry Olympiad, which enabled him to enter the academic path in chemistry, which he studied as a student at his home university, the University of Salamanca. Gómez-Bombarelli stayed on as a doctoral student, where he studied the effects of DNA-damaging chemicals.
“My PhD started with experiments, and about halfway through I got caught up in the simulation and computer science bug,” he says. “I started simulating the same chemical reactions I was measuring in the lab. I like the way programming organizes the brain. It seemed like a natural way to organize thinking. Programming is also much less limited by what you can do with your hands or scientific instruments.”
Gómez-Bombarelli then went to Scotland for a postdoctoral position, where he worked on quantum effects in biology. Through this work, he came into contact with Alan Aspuru-Guzik, a professor of chemistry at Harvard University, which he joined for another postdoctoral fellowship in 2014.
“I was one of the first people to use generative AI in chemistry in 2016, and in 2015 I was part of the first team to use neural networks to understand molecules,” says Gómez-Bombarelli. “These were the early beginnings of deep learning in science.”
Gómez-Bombarelli also began working to eliminate the manual parts of molecular simulations in order to conduct more high-throughput experiments. He and his colleagues performed hundreds of thousands of calculations on a variety of materials, discovering hundreds of promising materials for testing.
After two years in the lab, Gómez-Bombarelli and Aspuru-Guzik founded a general-purpose materials computing company that eventually focused on the production of organic light-emitting diodes. Gómez-Bombarelli joined the company full-time and says it was the hardest thing he's ever done in his career.
“It was amazing to create something tangible,” he says. “Moreover, after I saw Aspuru-Guzik running a lab, I didn't want to become a professor. My dad was a professor of linguistics and I thought it was a quiet job. Then I saw Aspuru-Guzik with a group of 40 people, who was on the road 120 days a year. It was crazy. I didn't think I had this kind of energy and creativity in me.”
In 2018, Aspuru-Guzik suggested Gómez-Bombarelli apply for a new position at MIT's Department of Materials Science and Engineering. However, fearing his job at the university, Gómez-Bombarelli missed the deadline. Aspuru-Guzik met him in his office, slammed his hands on the table and said, “You have to apply for this.” This was enough for Gómez-Bombarelli to submit a formal application.
Fortunately, at his startup, Gómez-Bombarelli spent a lot of time thinking about how to generate value from the discovery of computational materials. During the interview, he says, he was fascinated by the energy and collaborative spirit at MIT. He also began to appreciate research opportunities.
“Everything I did as a postdoc and in the company was intended to be a subset of what I could do at MIT,” he says. “I was building products and I still do that. Suddenly the world of my work became a subset of a new universe of things I could explore and do.”
Nine years have passed since Gómez Bombarelli joined MIT. Today, his laboratory focuses on the influence of atomic composition, structure and reactivity on the properties of materials. He also used high-performance simulations to create new materials and helped develop tools to combine deep learning with physics-based modeling.
“Physics-based simulations make data and AI algorithms better and better the more data you give them,” says Gómez Bombarelli. “There are a lot of virtuous cycles between AI and simulation.”
The research group he created is purely computational – it does not conduct physical experiments.
“It's a blessing because we can have a huge range of activities and do many things at once,” he says. “We love working with experimenters and strive to be good partners with them. We also love creating computational tools that help experimenters sort through ideas from AI.”
Gómez-Bombarelli continues to focus on real-world applications of the materials he invented. His lab works closely with companies and organizations, such as the MIT Industrial Connectivity Program, to understand the material needs of the private sector and the practical obstacles to commercial development.
Accelerating learning
As excitement around artificial intelligence exploded, Gómez-Bombarelli noticed that the field was maturing. Companies like Meta, Microsoft and Google's DeepMind now regularly run physics-based simulations similar to those he worked on in 2016. In November, the U.S. Department of Energy launched the Genesis Mission to accelerate scientific discovery, national security and energy dominance through artificial intelligence.
“AI for simulation has gone from something that maybe could work to a consensus-based science,” says Gómez-Bombarelli. “We are at a turning point. People are thinking in natural language, we are writing articles in natural language, and it turns out that these large language models that have mastered natural language have opened up the possibility of accelerating learning. We have seen that scaling works in simulation. We have seen that scaling works in language. Now we will see how scaling works in science.”
When he first came to MIT, Gómez-Bombarelli says he was amazed by the lack of competition between researchers. He tries to bring the same positive-sum thinking to his research group, which consists of about 25 undergraduate and graduate students.
“We naturally became a really diverse group with a diverse mentality,” says Gomez-Bombarelli. “Everyone has their own career aspirations and strengths and weaknesses. The fun is discovering how to help people become the best version of themselves. Now I'm the one pushing people to apply for faculty positions after the deadline. I guess I've passed the baton.”
