VenusFactory2 Uses Self-Evolving Agents for Autonomous Protein Discovery
VenusFactory2 framework demonstrates self-evolving multi-agent system for protein discovery from natural language prompts, outperforming existing agents on VenusAgentEval.
Protein scientific discovery is bottlenecked by manual orchestration of information and algorithms. VenusFactory2 provides an autonomous framework that shifts from static tool usage to dynamic workflow synthesis via a self-evolving multi-agent infrastructure (arXiv:2603.27303). It outperforms a set of well-known agents on the VenusAgentEval benchmark.
The system autonomously organizes the discovery and optimization of proteins from a single natural language prompt. This aligns with prior AI applications in structural biology, including AlphaFold for protein structure prediction (Nature, 2021, https://www.nature.com/articles/s41586-021-03819-2).
Earlier directed evolution methods, recognized by the 2018 Nobel Prize in Chemistry to Frances Arnold, relied on laboratory iteration. VenusFactory2 integrates these concepts into an autonomous AI loop (https://www.nobelprize.org/prizes/chemistry/2018/summary/).
VenusFactory2: Self-evolving AI agents successfully perform directed evolution for protein discovery from a single prompt, accelerating autonomous scientific breakthroughs by removing manual workflow design.
Sources (3)
- [1]Primary Source(https://arxiv.org/abs/2603.27303)
- [2]AlphaFold Nature Paper(https://www.nature.com/articles/s41586-021-03819-2)
- [3]Directed Evolution Nobel Summary(https://www.nobelprize.org/prizes/chemistry/2018/summary/)