MicroGrowAgents: Multi-Agent AI for Microbial Cultivation

Overview

MicroGrowAgents is an agent-based system for AI-driven microbial cultivation and growth media design. It bridges the microbial cultivation gap through AI-powered multi-agent systems that integrate knowledge graphs, machine learning, and experimental automation.

The Challenge: Designing growth media for novel or fastidious organisms is slow and largely manual. The knowledge needed — cultivation protocols, metabolic capabilities, chemical requirements — is scattered across literature, genomes, and culture collection databases, and no single model captures all of it.

The Solution: MicroGrowAgents coordinates specialized agents, each focused on one source of evidence (literature, cross-organism analogy, genome function, media formulation). Their outputs are combined into organism-specific, evidence-based media recommendations grounded in the kg-microbe knowledge graph. The RuleML/GOBLIN lecture describes it as a hierarchical agentic-AI framework of 100+ specialist agents; the four agent roles documented below are the ones described in detail.

Specialized Agents

📚 LiteratureAgent

Mines 245+ papers for cultivation protocols, extracting growth conditions and media compositions from the published record.

🔁 AnalogyReasoningAgent

Performs cross-organism comparison and reasoning, transferring cultivation knowledge from well-characterized organisms to related, less-studied taxa.

🧬 GenomeFunctionAgent

Detects auxotrophies from genome annotations — built on 57 Bakta-annotated genomes spanning 667K features — to predict which nutrients an organism cannot synthesize and therefore requires in its media.

🧪 MediaFormulationAgent

Produces schema-driven media recommendations with evidence-based ingredient suggestions, assembling the other agents’ findings into a concrete, formulatable recipe.

Key Achievements

• 864,363 validated species across bacteria, archaea, fungi, and protozoa (GTDB + LPSN + NCBI)
• Multi-modal reasoning combining literature mining, metabolic modeling (FBA / gap-filling), and chemical similarity (208K+ embeddings)
• Genome-guided design for organism-specific media formulation

Technical Architecture

Multi-Agent Reasoning Pipeline

Target Organism
    ↓
┌─────────────────────────────────────────────┐
│  LiteratureAgent   AnalogyReasoningAgent      │
│  GenomeFunctionAgent   MediaFormulationAgent  │
└─────────────────────────────────────────────┘
    ↓ (evidence integration over kg-microbe)
Organism-Specific Media Recommendation

Integration with the CultureBotAI Ecosystem

• Depends on:
  • kg-microbe - knowledge graph foundation
  • MicroMediaParam - chemical compound mappings
  • eggnogtable - genome functional annotations
  • MATE-LLM - literature extraction
• Feeds Into:
  • Media formulation recommendations
  • PFASCommunityAgents - consortium design
• Works With:
  • MicroGrowLink - complementary graph/transformer-based prediction approach

Repository & Documentation

• GitHub: github.com/CultureBotAI/MicroGrowAgents (private repository — public release planned)
• Preprint: MicroGrowAgents: An Agentic AI System for Microbial Cultivation Engineering (bioRxiv, 2026)
• License: BSD-3-Clause
• Languages: Python, HTML, Shell, R
• Topics: ai4curation · monarchinitiative

Related Tools

• CultureMech - Microbial culture media knowledge graph (10,000+ recipes)
• MediaIngredientMech - LLM-assisted ingredient ontology mapping
• CommunityMech - Microbial community interaction modeling
• TraitMech - Microbial ecophysiological trait knowledge base
• ProteinTraitsMech - Protein sequence, structure, and function traits
• MicroGrowLink - Graph-based growth media prediction
• kg-microbe - Central knowledge graph for microbial cultivation

Research Impact

MicroGrowAgents is part of the KG-Microbe knowledge graph ecosystem developed at Lawrence Berkeley National Laboratory. It supports:

• AI-driven media design for novel and fastidious organisms
• Genome-guided prediction of nutritional requirements
• Evidence-based cultivation protocol synthesis from literature
• Data-driven cultivation optimization

Applications: The RuleML/GOBLIN lecture reports high-throughput results in which MicroGrowAgents and KOGUT improved growth and rare-earth-element depletion in Methylorubrum extorquens AM1.

Citation: Naseem, S., Miller, M. A., Martinez-Gomez, N. C., Sun, N., & Joachimiak, M. P. (2026). MicroGrowAgents: An Agentic AI System for Microbial Cultivation Engineering. bioRxiv. 10.64898/2026.06.04.729985

See also the KG-Microbe publication in GigaScience for details on the broader knowledge graph ecosystem.

Contact & Collaboration

For questions about MicroGrowAgents or collaboration opportunities:

• Principal Investigator: Dr. Marcin P. Joachimiak
• Email: mjoachimiak@lbl.gov
• Organization: CultureBotAI
• Laboratory: Environmental Genomics and Systems Biology Division, Lawrence Berkeley National Laboratory

Bibliography

1. Naseem S, Miller MA, Martinez-Gomez NC, Sun N, Joachimiak MP. MicroGrowAgents: An Agentic AI System for Microbial Cultivation Engineering. bioRxiv. 2026. doi:10.64898/2026.06.04.729985
2. Santangelo BE, Hegde H, Caufield JH, Reese J, Kliegr T, Hunter LE, Lozupone CA, Mungall CJ, Joachimiak MP. KG-Microbe — Building Modular and Scalable Knowledge Graphs for Microbiome and Microbial Sciences. GigaScience. 2026;giag077. doi:10.1093/gigascience/giag077
3. Máša P, Kliegr T, Joachimiak MP. Explainable rule-based prediction of cultivation media for microbes. Computational and Structural Biotechnology Journal. 2025;27:5194–5206. doi:10.1016/j.csbj.2025.10.014 · free full text
4. Joachimiak MP. Knowledge Oriented Graph Unified Transformer (KOGUT) v0.1 [software]. DOE CODE; 2025. doi:10.11578/dc.20260210.3 · DOE CODE 175162
5. Joachimiak MP. “RuleML/GOBLIN COST Action Lecture on Data Science: Teaching AI to Teach Humans About Microbiology” [talk]. RuleML / COST GOBLIN Action Seminar; 2026. Recording

Full publication list →