Alex Rives

Alex Rives is head of science at Biohub, leading its work at the intersection of artificial intelligence and biology. He founded and led the Evolutionary Scale Modeling (ESM) project at Meta’s fundamental AI research lab and was a founder and chief scientist at EvolutionaryScale. His work centers on language models for proteins, with applications in structure prediction, protein design and understanding the effects of mutations.[3][4]

In a November 2025 interview, Mark Zuckerberg announced that the EvolutionaryScale team would join Biohub and that Rives would run the program, partnering with leading biologists. Zuckerberg described putting an AI researcher in charge as a reflection of his optimism that AI would be fundamental to the work. By May 27, 2026, Rives was Biohub’s head of science and announced ESMFold2.[2][1]

ESMFold2 builds on ESMC’s learned protein representations to predict structures. Its looped transformer passes representations through the same parameters repeatedly, refining them on each pass; training takes gradients through that recurrent process. Running more loops allows additional computation at inference time without retraining. The model can operate without explicit multiple sequence alignments, but optionally accepts them to improve accuracy.[6]

The Biohub team tested computationally designed binders against five oncology and immunology targets: EGFR, PDGFRβ, PD-L1, CTLA-4 and CD45. It generated candidate pools at two computational scales and screened the top 84 designs for each target, molecular format and scale. Biohub reported that higher-compute search raised average laboratory binding success from 54% to 70% for compact minibinders and from 12% to 21% for single-chain variable fragment antibodies. These results concern laboratory binding, rather than clinical efficacy.[6]

This work builds on the earlier ESM research program Rives led at Meta, which explored whether learning from protein sequences could yield useful representations of biological structure and function.[3][4][6]

Early ESM models learned to predict randomly masked amino acids from the surrounding sequence, using millions of protein sequences drawn from across life. The team found that their internal representations encoded structure and function without structural supervision, and that richer representations emerged with more training compute. Rives’s publication record includes Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences, alongside collaborators including Joshua Meier, Tom Sercu and Siddharth Goyal.[6][7]

At EvolutionaryScale, where Rives was a founder and chief scientist, the team extended this approach to generation. Tom Sercu’s June 25, 2024 ESM3 announcement described a transformer that jointly models protein sequence, structure and function. All three are tokenized and partially masked during training; the model predicts the missing positions. At generation time, it progressively fills masked positions, guided by partial or complete specifications across those three tracks.[3][5]

The team’s fluorescent-protein experiment made that generative capability concrete. In a first round of 96 candidates, it found B8, a protein distant from natural fluorescent proteins but 50 times less bright, with a chromophore that matured over a week. Starting from B8, another round of 96 candidates produced esmGFP, with brightness similar to natural GFPs. Sercu’s announcement reported 58% sequence similarity to the closest fluorescent protein found in nature. Its estimate of an equivalent to more than 500 million years of evolution came from natural GFP diversification rates.[5]

The ESM3 team released the weights and code for a 1.4-billion-parameter model under the MIT license, continuing the project’s commitment to open science. Biohub subsequently released the models in its ESMC and ESMFold2 protein release under the same license.[5][6]

Rives earned a bachelor’s degree in philosophy and biology from Yale University and a doctorate in computer science from New York University. He is an assistant professor in MIT’s Department of Electrical Engineering and Computer Science and a core institute member at the Broad Institute of MIT and Harvard, which lists him as on leave.[3][4]