Protein language models trained on diverse evolutionary sequences
ProGen 2 is a suite of protein language models trained on over 1 billion protein sequences across genomic and metagenomic databases for controllable generation and fitness prediction.
Trained on over 1 billion sequences from UniParc, BFD, and metagenomic sourcesControllable sequence generation with tag-based conditioningHigh accuracy zero-shot variant effect predictionModel sizes ranging from 150M to 6.4B parameters
ProGen 2 is a suite of protein language models trained on over 1 billion protein sequences across genomic and metagenomic databases for controllable generation and fitness prediction.
信息已比对权威文献或官方白皮书;非实验性跑分。 权威出处 · 最后核对: 20/09/2026, 11:09:40 UTC
核心功能特性
Trained on over 1 billion sequences from UniParc, BFD, and metagenomic sources
Controllable sequence generation with tag-based conditioning
High accuracy zero-shot variant effect prediction
Model sizes ranging from 150M to 6.4B parameters
Interactive 3D Illustration
Biological Architecture (Stylized Ribbon)
Default view is a stylized demonstration ribbon. Enter any 4-letter RCSB PDB accession code above to stream real atomic coordinates.
⇄ Drag to rotate · Scroll to zoom
Illustrative Confidence Gradient:
>90 High / Low B 70-90 Confident 50-70 Low <50 Very low / High B
Academic Context & Research Evidence
Biological & Workflow Fit
Biological Application
Enzyme engineering, antibody optimization, and zero-shot fitness prediction
Research Workflow
Generate sequence variants with high predicted stability and activity
Use this citation format when referencing ProGen 2 in scientific publications and benchmark papers.
@software{progen_2_2026,
title = {{ProGen 2}},
author = {{Salesforce AI Research}},
year = {2026},
url = {https://github.com/salesforce/progen},
note = {Indexed on aibioatlas - AI for Biology and Drug Discovery}
}
Peer-Reviewed Literature & Preprints
Live scientific citations queried from Europe PMC and PubMed for ProGen 2.
⏳ Querying Europe PMC & PubMed for indexed publications...
技术参数与产品规格
Missing values mean the catalog has no recorded information. They do not mean a feature is absent.
Entry typeAI Model
Access modeOpen Source
AI roleGenerative Protein Modeling
Input dataNot recorded
Output dataNot recorded
Licence conditionsBSD-3-Clause
Commercial eligibilityPermissive open source release
Compute requirements1x GPU (>= 16GB VRAM) for inference
ValidationNot recorded
TypeAutoregressive protein language model
Intended useNot recorded
CompatibilityNot recorded
ManufacturerSalesforce AI Research
Biological applicationEnzyme engineering, antibody optimization, and zero-shot fitness prediction
Research workflowGenerate sequence variants with high predicted stability and activity