agentic biotechnology workflows
Agentic Biotechnology Workflows (2025)
Section titled “Agentic Biotechnology Workflows (2025)”What they automate, bio domain, and how they connect to Claude/Codex/MCP.
1. Virtual Lab of AI agents
Section titled “1. Virtual Lab of AI agents”Virtual Lab of AI agents (Nature, Jul 29, 2025) — Multi-agent “PI/Scientist/Critic” workflow that proposes designs, selects tools (ESM/AF-Multimer/Rosetta), runs round-based optimization, and ships wet-lab-validated nanobodies; definitive proof that agents can drive ideation→in silico→in vitro with minimal hand-offs. LLM backend: not disclosed publicly in paper text; orchestration pattern is model-agnostic. Claude/Codex: not reported. MCP: not reported. Wet-lab: yes. Paper
2. CRISPR-GPT
Section titled “2. CRISPR-GPT”(Nat Biomed Eng, Jul 31, 2025) — Multi-agent genome-editing co-pilot (Planner/Executor/Tool-Provider) covering system selection, gRNA design, delivery, protocols, analysis; demonstrates AI-guided knockouts (Cas12a) and CRISPRa in human cell lines. LLM backend: GPT-4o reported. Claude/Codex: not reported. MCP: not reported. Wet-lab: yes. Paper
3. Agent Laboratory
Section titled “3. Agent Laboratory”(arXiv, v2 Jun 17, 2025) — Full research-process automation (literature→experiments→paper/code); best results with OpenAI o1-preview; reusable orchestration patterns (roles, review loops). Bio relevance: general, applicable to bio pipelines. Claude/Codex: not reported. MCP: not reported. Wet-lab: no (computational + human eval). Paper
4. BioMARS
Section titled “4. BioMARS”(arXiv, Jul 2, 2025) — Hierarchical multi-agent + modular robotics platform: Biologist agent (RAG protocol synthesis) → Technician agent (robotic pseudo-code) → Inspector agent (VLM anomaly detection). Targets autonomous planning and execution of biological experiments. Claude/Codex: not reported. MCP: not reported. Wet-lab: robotic execution focus (platform paper). Paper
5. Multi-Agent Drug Discovery & Clinical Simulation
Section titled “5. Multi-Agent Drug Discovery & Clinical Simulation”(bioRxiv, Aug 2025) — Proposes modular DMTA→PK/PD→trial simulation agent stack; surveys 51 LLM-based studies; emphasizes interoperability and human-in-the-loop checkpoints. Claude/Codex: not reported. MCP: not reported. Wet-lab: conceptual + survey. Paper
6. AgentD
Section titled “6. AgentD”(bioRxiv/arXiv, Jul 2025) — Modular drug-discovery agent for retrieval, molecule generation/refinement, ADMET, and 3D protein–ligand pose generation; describes task APIs and workflow glue. Claude/Codex: not reported. MCP: not reported. Wet-lab: computational. bioRxiv Paper | arXiv Paper
7. PharmaSwarm
Section titled “7. PharmaSwarm”(arXiv, Apr 2025) — MAS for hypothesis-driven drug discovery; containerized agents orchestrated via low-code or Kubernetes; integrates mechanistic sims & interpretable AI; strong software ops detail. Claude/Codex: not reported. MCP: not reported. Wet-lab: not in paper. Paper
8. Automating AI Discovery for Biomedicine
Section titled “8. Automating AI Discovery for Biomedicine”(bioRxiv, Jun 2025) — Agents over a biomedical knowledge-graph to plan studies and propose interventions; reports wet-lab validation of selected predictions. Claude/Codex: not reported. MCP: not reported. Wet-lab: yes (validation). Paper
9. OmicsNavigator
Section titled “9. OmicsNavigator”(bioRxiv, Jul 2025) — Multi-agent omics workflow (data QC→analysis→interpretation→hypothesis generation) with autonomous study distillation. Claude/Codex: not reported. MCP: not reported. Wet-lab: no (bioinformatics). Paper
10. Biomni
Section titled “10. Biomni”(bioRxiv, Jun 2025) — General-purpose biomedical agent executing a wide spectrum of research tasks with tool use and autonomy; positions as a lab-wide orchestrator. Claude/Codex: not reported. MCP: not reported. Wet-lab: not directly. Paper
11. SpatialAgent
Section titled “11. SpatialAgent”(bioRxiv, Apr 2025) — Autonomous agent for spatial biology (panel design, analysis, reporting) with dynamic tool execution and adaptive policies. Claude/Codex: not reported. MCP: not reported. Wet-lab: no. Paper
12. PharmAgents
Section titled “12. PharmAgents”(arXiv, Mar 2025) — “Virtual pharma” multi-agent ecosystem spanning target ID→lead optimization→toxicity and synthetic feasibility checks; emphasizes interpretability and self-evolution. Claude/Codex: not reported. MCP: not reported. Wet-lab: in silico focus. Paper
13. Multi-Agent Automated Platform for Scientific Discovery
Section titled “13. Multi-Agent Automated Platform for Scientific Discovery”(bioRxiv, Jul 2025) — Uses LLMs to design experimental interventions and auto-formalize hypotheses for lab automation integration. Claude/Codex: not reported. MCP: not reported. Wet-lab: platform-level. Paper
14. Steering towards safe self-driving laboratories
Section titled “14. Steering towards safe self-driving laboratories”(Nat Rev Chem, 2025) — Perspective mapping governance, safety patterns, and workflow frameworks (e.g., AlabOS) for autonomous labs; essential for risk/ops in agentic wet-lab stacks. Claude/Codex: N/A. MCP: N/A. Wet-lab: conceptual. Paper
15. Democratizing automation & the evolution toward true autonomous labs
Section titled “15. Democratizing automation & the evolution toward true autonomous labs”(RSC Chem. Sci., Aug 2025) — Reviews autonomous robotic lab capabilities and benefits across synthesis and materials/chem/bio discovery; ties together informatics+robotics+AI. Claude/Codex: N/A. MCP: N/A. Wet-lab: ecosystem review. Paper
16. ORGANA
Section titled “16. ORGANA”(Elsevier, 2025) — Robotic assistant for automated chemistry with decision-making & perception; relevant to drug-synthesis automation pieces of agentic pipelines. Claude/Codex: not reported. MCP: not reported. Wet-lab: robotic platform. Paper
17. STELLA
Section titled “17. STELLA”(arXiv/bioRxiv, Jul 2025) — Self-evolving biomedical research agent that expands domain skills over time; focuses on lifecycle learning and autonomy levels. Claude/Codex: not reported. MCP: not reported. Wet-lab: not shown. arXiv Paper | bioRxiv Paper
18. FROGENT
Section titled “18. FROGENT”(arXiv, ~Aug 2025) — End-to-end drug-design agent explicitly using the Model Context Protocol (MCP) to integrate biochemical DBs, extensible tool libraries, and task-specific AI; rare, explicit MCP adoption in drug discovery. Claude/Codex: not reported. MCP: yes (explicit). Wet-lab: in silico pipeline. Paper
19. Automated MD for Proteins
Section titled “19. Automated MD for Proteins”(arXiv, Jul 2025) — LLM-driven automation of protein molecular-dynamics setup via scripted agents; shows agent loops around simulation tooling. Claude/Codex: not reported. MCP: not reported. Wet-lab: no. Paper
20. ToolUniverse
Section titled “20. ToolUniverse”(Zitnik Lab, HMS, 2025) — Comprehensive platform providing 211+ biomedical tools accessible through Model Context Protocol (MCP) for “AI Scientist” agents; enables broad integration of biomedical tools and databases for automated research workflows. Claude/Codex: not reported. MCP: yes (native integration with 211+ biomedical tools). Wet-lab: tool integration platform. Project | GitHub
21. Agentic AI for Scientific Discovery — Survey
Section titled “21. Agentic AI for Scientific Discovery — Survey”(ICLR-adjacent, Mar 2025) — Taxonomy of agent workflows, evaluation metrics, and cross-domain toolchains including bio/chem; useful index of frameworks and datasets. Claude/Codex: not central. MCP: discussed per protocol landscape papers. Wet-lab: N/A. Paper