GENERATIVE MODEL01 ↗
⌬
ESM3
Multimodal protein generation conditioned on sequence, structure or functional information.
BEST FOR Protein design and guided generation
Learn ESM3 THE INDEPENDENT PROTEIN AI LEARNING HUB / 2026 EDITION
Explore the language
of life itself.
Learn how protein language models connect sequence, structure and function. Evidence-led explainers, practical ESM3 tutorials and the latest developments across the ESM ecosystem.
✳ Research grounded⌁ Open educational resources↗ Updated for 2026
01 / SEQUENCE
02 / STRUCTURE
03 / FUNCTION
Three perspectives. A new way to understand proteins.
PROTEIN DESIGN, DEMYSTIFIEDSCROLL TO EXPLORE
ESM3 AI / YOUR RESEARCH COMPANION
AI + source linksAsk ESM3 AI. Explore protein science.
Ask questions about ESM3, model selection and protein science. Answers refer to published Academy articles with source links.
What would you like to understand?
Ask a scientific question or choose a suggested prompt. Results cite the Academy library; verify critical details against primary sources.
Educational information, not laboratory or clinical advice. Confirm details against primary sources.
Need help choosing a model? Open the interactive selector
01 / What would you like to do?
YOUR RECOMMENDED STARTING POINT
ESM3 ↗
Explore controlled protein generation using sequence, structure and function inputs.
Computational outputs require task-specific and experimental validation.
01 / START YOUR JOURNEY
From curious to capable.
Explore a learning path built around real questions, transparent assumptions and workflows you can follow.
FOUNDATIONS 01
Understand the science
Discover what ESM3 is, why multimodal protein modeling matters and where it stops.
Explore fundamentals ↗ PRACTICE 02Build your first workflow
Install tools, run inference, read outputs and avoid outdated instructions.
Browse tutorials ↗ RESEARCH 03Go deeper, responsibly
Compare models, evaluate predictions and plan reproducible experiments.
Explore deep dives ↗02 / THE ESM ECOSYSTEMDifferent models.
Different models.
Different questions.
No single model does everything. Choose the right tool for your scientific question.
REPRESENTATION MODEL02 ↗
∿
ESMC
Protein sequence representations for embeddings, analysis and downstream learning.
BEST FOR Sequence representation and inference
Explore ESMC STRUCTURE MODEL03 ↗
◈
ESMFold2
High-resolution structure predictions from sequence, including complex modeling.
BEST FOR Structural prediction and complexes
Explore ESMFold2 Model descriptions based on official Biohub documentation. Computational predictions are hypotheses, not experimental proof.
COMPLETE LIBRARY / ORIGINAL STRUCTURE
Explore every Academy subject.
The eight original top-level subjects remain. Updated historical URLs are retained for the articles currently available.
About ESM3Models, concepts and starting points↗Use CasesBiology, materials and research scenarios↗TutorialsPractical guided workflows↗Advanced TechniquesScaling, evaluation and research methods↗Best PracticesReproducibility, data care and ethics↗Developer ToolsSDKs, APIs and integrations↗FAQs & Help CenterSetup, compatibility and troubleshooting↗Community NewsResearch updates and ecosystem changes↗
04 / FROM THE LIBRARY
Useful ideas, carefully explained.
Practical guides that connect protein modeling concepts to reproducible research.
BEGINNER TUTORIALS
Installing and Configuring ESM3: Step-by-Step
Updated ESM3 installation guide: pip, Biohub API access, local model requirements, troubleshooting and links to the official repository.
Read guide ↗BEGINNER TUTORIALS
Running Your First ESM3 Model: A Hands-On Tutorial
Learn the supported ESMProtein and GenerationConfig interfaces with a small educational sequence-completion exercise.
Read guide ↗INTERMEDIATE TUTORIALS
Fine-Tuning ESM3 for Custom Applications
A practical decision guide to adapting ESM3: prompts, supported training workflows, evaluation splits, compute and licensing.
Read guide ↗PROTEIN MODELING AND STRUCTURAL BIOLOGY
Protein Structure Prediction Using ESM3
Understand why structure generation with ESM3 and structure prediction with ESMFold2 solve different scientific tasks.
Read guide ↗BIOMEDICAL APPLICATIONS
Utilizing ESM3 for Drug Discovery Research
Research-oriented overview of how ESM3 can support protein design hypotheses while separating computational scores from clinical evidence.
Read guide ↗ADVANCED TECHNIQUES
Scalable Training Methods for Large ESM3 Models
Understand scalable ESM inference, experiment tracking and the limits of undocumented distributed-training recipes.
Read guide ↗05 / SCIENCE WITH CONTEXT
Extraordinary claims deserve extraordinary clarity.
ESM3 generated a fluorescent protein that was experimentally characterized in the original research. The widely quoted “500 million years” is a comparative evolutionary estimate — not a literal computer simulation of 500 million years.
Explore the peer-reviewed study ↗06 / LEARN WITH PURPOSE
Get in touch ↗Keep asking better questions.
Have a correction, research question or collaboration idea? Help us make protein AI education more reliable and accessible.