Rafa Gómez-Bombarelli

Rafa Gómez-Bombarelli, also known as Rafael Gómez-Bombarelli, works at the intersection of machine learning and chemistry. He is Lila Sciences’ chief scientific officer for physical sciences and joined MIT’s materials science and engineering faculty in 2018. His research combines physics-based atomistic simulation with machine learning to connect materials’ atomic composition, structure and reactivity to their performance.[1][2][3]

As a Lila co-founder, Gómez-Bombarelli leads AI strategy and execution for chemistry and materials, working across experimental data and physics-based simulations. Scientific superintelligence is the stated goal of that work.[2]

His public discussion of AI also emphasizes the constraints of making useful materials. The companion article for his July 2026 Latent Space interview with Andy Beam summarizes his “bittersweet lesson”: while scaling provides a roadmap in AI, it acts as a filter in materials science, because only things that can scale ultimately matter.[1]

His earlier work provides concrete examples of using learned representations to design molecules and simplify physical simulations. Before Lila, his career moved through experimental chemistry, computational screening, commercial OLED development and MIT research.[3][5][6]

Gómez-Bombarelli earned his BS, MS and PhD at the University of Salamanca in Spain, then undertook postdoctoral research at Heriot-Watt University and Harvard. In Alán Aspuru-Guzik’s Harvard laboratory, he worked on high-throughput virtual screening for organic light-emitting diodes (OLEDs) and battery electrolytes. In 2016, he joined Kyulux as a senior researcher, applying Harvard-licensed technology to commercial OLED products.[3]

In Automatic chemical design using a data-driven continuous representation of molecules, first submitted in October 2016, Gómez-Bombarelli and a team including Jennifer N. Wei, David Duvenaud and Alán Aspuru-Guzik coupled three neural-network functions: an encoder converts a molecular representation into a continuous vector, a decoder converts vectors back into molecular representations, and a predictor estimates chemical properties from those vectors. This makes molecular search possible through operations such as interpolation, perturbation and gradient-based optimization. The authors reported training on hundreds of thousands of structures and demonstrations on drug-like molecules and molecules with fewer than nine heavy atoms.[5]

With Wujie Wang, he developed Autograin, first submitted in December 2018. Coarse-graining reduces a simulation’s dimensionality and averages out fast motions, allowing larger systems and longer timesteps. Autograin uses autoencoders to learn both the mapping from all-atom coordinates to a reduced representation and the corresponding energy model. Reconstruction loss helps learn the reduced variables; force matching determines the coarse-grained potential-energy function. Wang and Gómez-Bombarelli reported tests on model systems, including single-molecule and bulk-phase periodic simulations.[6]

After joining MIT in 2018, Gómez-Bombarelli pursued materials discovery through physical chemistry and machine learning. His group’s work includes designing zeolite templates to control pores that trap selected molecules, and developing machine-learning approaches to identify photoswitches—molecules activated by light—with potential applications in light-activated drugs.[3]

He also co-founded Calculario, a materials-discovery company using quantum chemistry and machine learning, and served as ZebiAI’s chief learning officer. When Relay Therapeutics announced its acquisition of ZebiAI in April 2021, he expressed the belief that combining Relay’s computational and experimental approaches with ZebiAI’s experimental datasets and machine-learning capabilities would “push the boundaries of what is possible in drug discovery.” The announcement said he would join Relay as an advisor; the transaction remained subject to closing conditions.[3][7]

At MIT, Gómez-Bombarelli co-created Machine Learning for Molecular Engineering with Conor Coley and Ernest Fraenkel, teaching applications across biological, chemical and materials engineering. The trio received the inaugural Common Ground Award for Excellence in Teaching in 2023, the same year Gómez-Bombarelli became an associate professor without tenure and a Sloan Research Fellow. MIT announced his tenure in May 2025.[4]