Platform · arXiv 2026

VenusFactory2

Self-evolving AI agents for protein discovery and directed evolution — a free web platform that orchestrates data, tools, and Venus / Prot* models in the browser.

Yang Tan, Lingrong Zhang, Mingchen Li, Yuanxi Yu, Bozitao Zhong, Bingxin Zhou, Nanqing Dong, Liang Hong

Web demo Tech report Code

What VenusFactory2 adds

The original VenusFactory (ACL Demo 2025) packaged data retrieval and PLM fine-tuning as an open toolkit. VenusFactory2 turns that stack into a free web product with playgrounds, APIs, and agent-oriented interfaces — so wet-lab and computational users can score mutants, run benchmarks, and call models without assembling local environments.

  • Browser access at venusfactory.bio
  • Interactive playground for fitness / engineering tasks
  • Integration of VenusX, VenusREM, ProSST, ProtSSN and related checkpoints
  • Technical report: arXiv:2603.27303

Integrated methods

Models and benchmarks wired into the platform include:

  • VenusX — fine-grained functional benchmark
  • VenusREM — retrieval-enhanced mutation scoring
  • ProSST / ProtSSN — sequence–structure PLMs
  • VenusRAR — agentic rank-and-reason selection (companion line)

Citation

If you use VenusFactory2, please cite the technical report:

@misc{tan2026venusfactory2,
  title={Self-evolving AI agents for protein discovery and directed evolution},
  author={Tan, Yang and Zhang, Lingrong and Li, Mingchen and Yu, Yuanxi and Zhong, Bozitao and Zhou, Bingxin and Dong, Nanqing and Hong, Liang},
  year={2026},
  eprint={2603.27303},
  archivePrefix={arXiv},
  primaryClass={q-bio.QM},
  url={https://arxiv.org/abs/2603.27303}
}

For the ACL Demo toolkit paper, see the VenusFactory page.