Protein Language Models and Generative Protein Design
Language and diffusion models over protein sequences and structures that generate new proteins, binders and antibodies.
Tier Grades
Tier A: Frontier AI labs
feasible
EvolutionaryScale trained ESM3 with over 1x10^24 FLOPS.
Tier B: Indonesian AI labs
not feasible
Pretraining at 1x10^24 FLOPS is out of reach; use of open models is possible but no Indonesian actor is observed.
Tier C: Institution units such as ITB
stretch
Running open RFdiffusion and ProteinMPNN fits a lab with wet-lab validation, which is not shown.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
not feasible
Protein design needs wet-lab testing outside delivery scope.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeESM3: 98B parameters trained with over 1x10^24 FLOPS, which EvolutionaryScale calls the most compute applied to a biological model.
- DataESM3: a multimodal dataset augmented with hundreds of millions of synthetic predicted structures and functions.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): Not in the sources read (the ESM3 blog gives no author count).
- Time horizonMulti-year.
- Shelf-life signalsESM3 is a generative model over sequence, structure and function together, beyond sequence-only protein language models.
- Pipeline roleTier A trains protein models; Tier C designs and tests proteins with open ones; Tier B and Tier D have no observed role.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
ESM3scientificexemplarRFdiffusionscientificexemplarProteinMPNNscientificexemplar
Chai-2scientificproduced by
Dayhoff-3b-GR-HM-cscientificproduced byESM Cscientificproduced byESM3 98Bscientificproduced byESM3 open 1.4Bscientificproduced by
BoltzGenscientificproduced by
ESM-2 / ESMFoldscientificproduced byEvoDiffscientificproduced byLinks
R&D TopicProtein Language Models and Generative Protein Design
ModelsESM3EvolutionaryScale, scientificRFdiffusionscientificProteinMPNNscientificBoltzGenBoltz PBC, scientific+7 more
ESM3EvolutionaryScale, scientificRFdiffusionscientificProteinMPNNscientificBoltzGenBoltz PBC, scientific+7 moreWho Runs It
PfizerContract signed, 8 Jan 2026
PfizerContract signed, 8 Jan 2026Models 11
ESM CEvolutionaryScale, scientificproduced byESM-2 / ESMFoldMeta, scientificproduced byESM3EvolutionaryScale, scientificexemplarESM3 98BEvolutionaryScale, scientificproduced byESM3 open 1.4BEvolutionaryScale, scientificproduced byBoltzGenBoltz PBC, scientificproduced byChai-2Chai Discovery, scientificproduced byDayhoff-3b-GR-HM-cMicrosoft AI, scientificproduced byEvoDiffMicrosoft AI, scientificproduced byProteinMPNNscientificexemplarRFdiffusionscientificexemplarPapers
- ProGen: Language Modeling for Protein GenerationAli Madani et al., arXiv preprint, 2020 founding
- ProtTrans: Towards Cracking the Language of Life's Code Through Self-Supervised Deep Learning and High Performance ComputingAhmed Elnaggar et al., arXiv preprint, 2020 founding
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- Evolutionary Scale , ESM3: Simulating 500 million years of evolution with a language model evolutionaryscale.ai https://www.evolutionaryscale.ai/blog/esm3-release
- raw.githubusercontent.com/RosettaCommons/RFdiffusion/main/README.md https://raw.githubusercontent.com/RosettaCommons/RFdiffusion/main/README.md
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Protein Language Models and Generative Protein Design", https://aurion.semai.id/topics/protein-language-models-and-generative-protein-design/, records of October 2026.