Alex Rives

Career

Rives started and led the Evolutionary Scale Modeling project at Meta's fundamental AI research lab, which built transformer language models trained on protein sequences. Rives was then a founder and the chief scientist of EvolutionaryScale, a public-benefit company developing AI for biology.¹

EvolutionaryScale is now part of Biohub, and its ESM models and services are offered through Biohub. As Head of Science, Rives leads Biohub's work combining artificial intelligence and biology.¹²³

Research

Rives was the first author of the paper showing that biological structure and function emerge from unsupervised learning on 250 million protein sequences, the work behind the ESM-1b model. Later work led to ESM-2 and ESMFold, which predicts protein structure directly from a sequence using the ESM-2 language model.⁴

The ESM3 preprint, with Rives as senior author, presents a model that reasons jointly over protein sequence, structure and function. EvolutionaryScale used ESM3 to design esmGFP, a new green fluorescent protein generated with chain-of-thought prompting.⁵²

Work and contributions

• ESM3 Open weights

  Senior author of the 2024 ESM3 preprint, which describes a generative language model that reasons over protein sequence, structure and function; the open small ESM3 model is released with source code on GitHub.⁵²³

• ESM-2 and ESMFold Open weights

  Senior author of the paper introducing ESM-2 and ESMFold, released through the Meta FAIR protein team's MIT-licensed ESM repository, which pairs the ESM-2 language model with end-to-end structure prediction from a single sequence.⁴

• ESM Cambrian Open weights

  EvolutionaryScale introduced ESM Cambrian in 2024 as a model family for protein representation learning, released at 300M, 600M and 6B parameters under the MIT license.⁶

• ESM3 open small (esm3-sm-open-v1) Open weights

  Last-listed author of the ESM3 paper in Science (doi:10.1126/science.ads0018), which the ESM3 README in the Biohub/esm repository asks users to cite.⁷

Timeline

1. Dec 2020The ESM repository added the ESM-1b protein language model described in Rives et al.⁴
2. Aug 2022The ESM-2 protein language models were released through the ESM repository.⁴
3. Nov 2022ESMFold, an end-to-end structure prediction model built on ESM-2, was released.⁴
4. 2024EvolutionaryScale introduced the ESM Cambrian protein representation models.⁶
5. Jul 2, 2024Posted the ESM3 preprint on simulating protein evolution with a language model as senior author.⁵

Other documented roles

Roles the sources document that are not confirmed as current.

• Founder and chief scientist, EvolutionaryScale¹²
• Started and led the Evolutionary Scale Modeling (ESM) project, Meta Fundamental AI Research¹⁴

Where they work

• Biohub 2 people in the wiki
• Broad Institute 1 person in the wiki
• MIT 26 people in the wiki

Related people

• Priscilla ChanAlso at Biohub
• Sergey EdunovLanguage models · AI for science
• Liam FedusLanguage models · AI for science
• Hoifung PoonAI for science · Language models
• Brian HieAI for science · Language models
• Jakob UszkoreitLanguage models · AI for science
• David WaddenLanguage models · AI for science
• Garyk BrixiAI for science · Language models
• Yejin ChoiLanguage models · AI for science
• Ali MadaniAI for science · Language models