Resources & Tools β€” KG-Microbe Knowledge Graph and AI Tools

CultureBotAI led by Dr. Marcin P. Joachimiak develops and maintains various computational resources, databases, and tools for the microbial research community, including the comprehensive KG-Microbe knowledge graph.

Quick Navigation

New to CultureBotAI? Start with Project Ecosystem & Workflows to understand how tools work together.

Looking for specific tools?

Want to see workflows? Jump to Common Workflows

🧬 KG-Microbe: Microbial Knowledge Graph

Overview

KG-Microbe is our flagship resource developed by Dr. Marcin P. Joachimiak - a comprehensive knowledge graph that integrates diverse microbial data sources to enable AI-driven insights and predictions.

πŸ“„ Read the Publication - GigaScience article detailing kg-microbe development and applications.

πŸ“‡ KG-Registry entry - Registry record with distributions and metadata. The graph carries 3,000+ organismal and 30,000+ genomic traits.

πŸ“‹ METPO Ontology Integration

The Microbial Ecophysiological Trait and Phenotype Ontology (METPO) plays a crucial role in kg-microbe by providing standardized terminology for microbial phenotypes and ecological characteristics.

Key Benefits:

  • Knowledge Organization - METPO terms provide semantic structure to organize diverse microbial data within the kg-microbe knowledge graph
  • Text Extraction - Standardized ontology terms power automated literature mining and text extraction processes
  • Semantic Consistency - Ensures consistent representation of microbial characteristics across different data sources

Links:

Key Features

  • Multi-source integration from major biological databases
  • Semantic consistency through ontology-driven organization
  • Machine-readable formats (RDF, Neo4j, JSON-LD)
  • Regular updates with automated data refresh pipelines
  • API access for programmatic data retrieval

Data Sources

kg-microbe integrates data from:

  • NCBI Taxonomy - Microbial taxonomy and phylogeny
  • UniProt - Protein sequences and functional annotations
  • GO (Gene Ontology) - Functional gene classifications
  • Environmental ontologies - Habitat and growth condition data
  • Literature sources - Manually curated cultivation data

Applications

  • Growth condition prediction for uncultured organisms
  • Taxonomic relationship exploration and phylogenetic analysis
  • Literature mining for cultivation protocols
  • Cross-organism comparison of growth preferences

Getting Started

# Clone the repository
git clone https://github.com/Knowledge-Graph-Hub/kg-microbe.git

# Install dependencies
cd kg-microbe
pip install -r requirements.txt

# Build latest knowledge graph
kg download
kg transform
kg merge

🌐 Project Ecosystem & Workflows

Understanding the CultureBotAI Ecosystem

The CultureBotAI toolkit consists of interconnected projects organized into a data processing pipeline with kg-microbe as the foundational knowledge graph.

Architecture Overview

                         β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
                         β”‚   kg-microbe    β”‚
                         β”‚  (Foundation)   β”‚
                         β””β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                                  β”‚
         β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
         β”‚                        β”‚                        β”‚
  β”Œβ”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”      β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”    β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”€β”€β”
  β”‚ Data        β”‚      β”‚ Chemical          β”‚    β”‚ Genome          β”‚
  β”‚ Ingestion   β”‚      β”‚ Processing        β”‚    β”‚ Analysis        β”‚
  β”‚             β”‚      β”‚                   β”‚    β”‚                 β”‚
  β”‚ β€’ assay-    β”‚      β”‚ β€’ CultureMech     β”‚    β”‚ β€’ eggnog_runner β”‚
  β”‚   metadata  β”‚      β”‚ β€’ MicroMediaParam β”‚    β”‚ β€’ eggnogtable   β”‚
  β”‚ β€’ MATE-LLM  β”‚      β”‚                   β”‚    β”‚                 β”‚
  β””β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”˜      β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜    β””β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”˜
         β”‚                       β”‚                        β”‚
         β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                                 β”‚
                    β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
                    β”‚   AI Agent Systems      β”‚
                    β”‚                         β”‚
                    β”‚ β€’ MicroGrowAgents       β”‚
                    β”‚ β€’ MicroGrowLink         β”‚
                    β”‚ β€’ PFASCommunityAgents   β”‚
                    β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                                 β”‚
         β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
         β”‚                       β”‚                       β”‚
  β”Œβ”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”     β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”   β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”€β”€β”
  β”‚ Specialized β”‚     β”‚ Web Services      β”‚   β”‚ Analysis        β”‚
  β”‚ Apps        β”‚     β”‚                   β”‚   β”‚ Tools           β”‚
  β”‚             β”‚     β”‚ β€’ MicroGrowLink   β”‚   β”‚                 β”‚
  β”‚ β€’ PFAS-AI   β”‚     β”‚   Service         β”‚   β”‚ β€’ microbe-rules β”‚
  β”‚ β€’ CMM-AI    β”‚     β”‚                   β”‚   β”‚                 β”‚
  β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜     β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜   β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

πŸ€– AI Curation Tools

The X-Mech Suite brings together sixteen ontology-grounded autonomous knowledge factories: HabitatMech, CommunityMech, TaxonMech, TraitMech, PathwayMech, CellStructureMech, ProteinTraitsMech, DUFMech, CofactorMech, NaturalProductMech, AntibioticMech, CMMMech, AssayMech, MediaIngredientMech, CultureMech and AIMech. Their curation workflows turn microbial-science and AI evidence into structured, machine-readable records. The suite overview and relationship graph show their shared vocabularies and documented connections. The culturebotai-claw manifest declares all sixteen members. The figures below were refreshed on November 9–10, 2026; DUFMech seed families, AIMech imported concepts and maintained records retain their distinct meanings.

Pipeline Overview

Raw Cultivation Records (Literature, Lab Protocols)
    ↓
CultureMech ↔ MediaIngredientMech ── candidate ingredient mappings ──→ KG-Microbe Knowledge Graph
Recipes       Ingredient identity                                          β”‚
    ↓                                                                      β”‚ ontologies, mappings,
CommunityMech + other domain-specific Mechs  β†β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ embeddings
Community, trait, taxon, habitat and molecular evidence
    ↓
AI Predictions (MicroGrowAgents, MicroGrowLink), drawing on the Mechs and KG-Microbe

CultureMech - Autonomous Knowledge Factory for Culture Media

Dedicated Page | GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code

15,915 curated culture media recipes from major international repositories, deduplicated into 6,325 canonical media, with LinkML schema, ingredient ontology grounding, and browser-based exploration.

What it does: Curates source-specific recipes, grounds ingredient identifiers, validates records, and generates deduplicated media and browser outputs.

β†’ Learn more on the dedicated CultureMech page

AIMech

GitHub Repository | Web Interface | CC BY 4.0 AIO imports Β· BSD-3-Clause code; upstream terms apply

AIMech extends the suite’s curation workflow to artificial intelligence concepts. Its pinned repository contains 1,929 concept records: 442 imported AIO classes and 1,487 maintained concepts with claim-specific citations and explicit AIO mappings or documented no-match assessments. Imported records are NOT_REVIEWED and maintained concepts are DRAFT; separate scientific review artifacts do not change those statuses. The repository provides native LinkML schemas and validation. Its data-license notice covers the AIO imports under CC BY 4.0; it does not specify a grant for maintained concepts. Code is BSD-3-Clause. Its website browses the current imported and maintained concepts by kind and AIO branch, with a page per concept.

CMMMech

GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code; upstream terms apply

CMMMech curates evidence-backed microbial extraction, transformation and recovery mechanisms for critical materials. Its 708 material records cover 457 mineral species, 111 secondary resources, 66 elements, 36 commodities, 31 primary resources and seven mineral groups. 500 records describe 609 mechanisms with cited evidence, assay context and limitations; 208 remain identity and characterization records without assigned microbial mechanisms. Its website browses its current records by material kind and by microbial process; the records, the LinkML schema and review artifacts are in the repository.

AssayMech

GitHub Repository | Records at Pin | CC BY 4.0 data Β· BSD-3-Clause code; upstream terms apply

AssayMech curates microbiological diagnostic and operational analytical assays, connecting METPO assay concepts to operational protocols, readouts, controls and limitations. Its pinned repository contains 9 assay records across four categories: six METPO-grounded diagnostic assay concepts and three operational analytical assays (ferrozine Fe(II), TEM-EDS intracellular bismuth and ICP-MS neodymium bioleaching) linked through four MEASURES_PROCESS cross-references to four CMMMech material records. No public GitHub Pages browser is deployed yet.

MediaIngredientMech - LLM-Assisted Ingredient Curation

Dedicated Page | GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code

2,956 curated ingredient records, 2,611 of them mapped (88% coverage), with LLM-assisted workflows for standardizing microbial cultivation ingredient data. Uses Large Language Models to intelligently map ingredient names to standardized ontology terms.

What it does: Curates ingredient identity, ontology mappings (mostly ChEBI, then MeSH, NCIT, MicrO, FOODON and ENVO; where curation found no term, about 175 kg-microbe registry ids and about 85 CAS numbers, plus about 55 placeholders pending curation), ENVO environmental context, and provenance through validated workflows with human oversight. Scoped updates preserve MIM-owned curation; see the dedicated page for supported commands.

β†’ Learn more on the dedicated MediaIngredientMech page

CommunityMech - Microbial Community Interaction Modeling

Dedicated Page | GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code

512 curated communities across 16 categories, modeled in LinkML with evidence-based ecological interactions for consortium design and multi-organism cultivation.

What it does: Provides structured representation of community composition, syntrophic interactions, and cultivation requirements for multi-species systems.

β†’ Learn more on the dedicated CommunityMech page

TraitMech - Microbial Ecophysiological Traits

GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code

Autonomous knowledge factory for microbial ecophysiological traits, seeded from METPO and curated incrementally β€” 1,130 trait records across 10 categories; 427 are marked REVIEWED and 769 carry causal graphs.

What it does: Standardizes the trait vocabulary used to describe microbial growth and ecology, and links traits to their evidence and, where a match exists, to kg-microbe.

PathwayMech - Microbial Pathway Mechanisms

GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code

Curates 282 pathway-level microbial mechanisms across 105 taxa, with explicit participants, reactions, 14,838 causal edges, and reference-backed evidence. Its KGX graph and SSSOM source mappings support reuse of pathway assertions and normalized identifiers. See the repository overview.

ProteinTraitsMech - Protein Sequence & Structure Traits

GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code (upstream sources retain their own terms)

Autonomous knowledge factory for protein sequence, structure, and function traits β€” 429,421 LinkML-validated records grouped under 34 browser source labels, one YAML per trait. These labels include imported sources and curation groupings; most records are imported and not yet reviewed. Reviewed records carry evidence-backed causal graphs.

What it does: Extends trait curation from the organism level to the molecular level, connecting protein features to the phenotypes they help explain. Explore the corpus, protein and ESM-2 sequence maps.

DUFMech - Domains and Families of Unknown Function

GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code

Provides a seed worklist of 8,296 Pfam families: 6,127 unknown candidates, 373 historical DUFs and 1,796 EX_DUF families renamed away from former DUF/UPF names. Family identifiers are measured alongside the other Mechs in the vocabulary heatmap and shared-term graph. Explore the worklists in the repository; the published browser now offers individual family pages, search, filters, evidence, curation history and cross-Mech leads. Native family records validate against a LinkML schema. Seed labels describe frozen metadata and naming history, not functional validation; all 8,296 families remain unscored.

CofactorMech - Microbial Enzyme Cofactors

GitHub Repository | Records at Pin | CC BY 4.0 data Β· BSD-3-Clause code; upstream terms apply

Curates microbial enzyme cofactors, coenzymes, prosthetic groups and catalytic metal ions grounded in ChEBI. Its pinned repository contains 14 cofactor records: six coenzymes, five metal ions and three prosthetic groups, with 57 cross-corpus links connecting cofactors to 25 MediaIngredientMech ingredients, 17 ProteinTraitsMech cofactor-requirement and enzyme records, and 10 PathwayMech pathways while preserving conjugate-form and precursor distinctions. No public GitHub Pages browser is deployed yet.

CellStructureMech - Microbial Cell Structures

GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code (redistributed UniProt and Complex Portal material retains its CC BY 4.0 terms)

1,022 structure records across 13 categories β€” organelles, envelope layers, appendages, microcompartments and multi-protein complexes β€” 873 of them grounded in GO cellular component.

What it does: Occupies the layer between traits and proteins, recording what a structure is made of, which organisms have it, what it does, and the causal mechanism by which it does so.

Related: Confers phenotypes recorded as TraitMech terms; hands single proteins off to ProteinTraitsMech.

HabitatMech - Microbial Habitats

GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code

3,209 habitat records harmonized from GOLD, BacDive, PREGO and Madin et al., grounded in ENVO, UBERON, FOODON, BTO and PO, with every source’s own attestation retained.

What it does: Gives the fleet one record per habitat concept, so that isolation sources expressed differently by each upstream database resolve to a single identity.

Related: Its causal graphs reuse most of TraitMech’s node types; TaxonMech leaves a taxon’s habitats and isolation sources to it; its site generator became AntibioticMech’s. See the relationship graph.

AntibioticMech - Antimicrobial Structures

GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code

3,125 antimicrobial chemical structures, 2,782 ontology-grounded, harmonizing ChEBI’s antimicrobial roles with CARD/ARO molecules, targets and resistance determinants.

What it does: Records one entry per antimicrobial structure, carrying its mode of action, molecular targets and the evidence for both where sources or curation have supplied them β€” 566 records have a mode of action and 378 a molecular target so far β€” and places all 3,125 on a chemical map by molecular fingerprint.

NaturalProductMech - Natural Product Structures

GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code

3,136 natural product structures, one per Standard InChIKey, seeded from ten sources and grounded in ChEBI, MIBiG and NCBI Taxonomy.

What it does: Records biosynthetic gene-cluster links for 3,123 structures and producer organisms for 3,097, and carries cited occurrences (2,351) and measured bioactivities (174) where a source reports them. Producer claims are graded: of 3,441, only 840 rest on evidence that addressed the organism. Of the 3,136 records, 20 are marked REVIEWED and 3,116 SEEDED; mechanism and evidence coverage vary by record.

TaxonMech - Microbial Taxa and Strains

GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code (upstream sources retain their own terms)

626,048 taxon records at species level and below, with 100,745 distinct listed strains, resolving NCBI Taxonomy, GTDB, LPSN and BacDive onto one NCBI-grounded identity.

What it does: Serves as the taxonomic counterpart of the other Mechs. Its 21,071 listed strains with genome links carry NCBI, GTDB, BV-BRC/PATRIC, IMG or AllTheBacteria identifiers; taxon pages also show StrainInfo references and the supporting evidence. Higher taxa appear only in a record’s lineage, never as records of their own.

Related: Leaves a taxon’s traits to TraitMech, its habitats to HabitatMech and its growth media to CultureMech, keeping only attestation counts. Most taxa the other Mechs cite are TaxonMech records, which the relationship graph shows as shared NCBI Taxonomy identifiers.

Common Workflows

Workflow 1: Novel Organism Media Prediction

  1. Start with organism taxonomy/genome
  2. Run eggnog_runner + eggnogtable (functional annotation)
  3. Query kg-microbe (related organisms, known preferences)
  4. Use MicroGrowAgents (integrate genome, literature, analogies)
  5. Get media recommendations with evidence

Workflow 2: Chemical Compound Knowledge Graph Integration

  1. Start with media composition text
  2. Run CultureMech (extract chemical entities)
  3. Run MicroMediaParam (map to ChEBI/PubChem)
  4. Integrate into kg-microbe (standardized chemical data)
  5. Enable downstream media predictions

Workflow 3: PFAS Biodegradation Consortia Design

  1. Query PFAS-AI database (identify candidate organisms)
  2. Extract genome features (eggnog_runner/eggnogtable)
  3. Query kg-microbe (environmental compatibility)
  4. Use PFASCommunityAgents (design optimized consortia)
  5. Get consortium composition + rationale

Workflow 4: Literature-Driven Culture Optimization

  1. Run MATE-LLM (extract cultivation protocols from papers)
  2. Integrate into kg-microbe (structured cultivation data)
  3. Use MicroGrowAgents LiteratureAgent (mine similar organisms)
  4. Get evidence-based media recommendations

Getting Started Guide

For Growth Media Prediction:

  • Start with: MicroGrowAgents or MicroGrowLink
  • Prerequisites: Access to kg-microbe knowledge graph
  • Recommended workflow: Workflow 1

For Chemical Data Processing:

  • Start with: CultureMech or MicroMediaParam
  • Prerequisites: Media composition text data
  • Recommended workflow: Workflow 2

For Specialized Research:

  • PFAS biodegradation: Start with PFAS-AI, then PFASCommunityAgents
  • Lanthanide bioprocessing: Start with CMM-AI

For Web-Based Access:

  • API users: Start with MicroGrowLinkService
  • Prerequisites: HTTP client, REST API knowledge

πŸ”§ CultureBotAI Software & Tools

Growth Media Prediction & Design

Private repository β€” public release planned | Python

Knowledge graph-based framework for predicting microbial growth media using advanced graph and transformer models. Integrates microbial, chemical, and environmental data into a heterogeneous knowledge graph and applies link prediction to forecast which media enable growth of given taxa.

Supported Models:

  • RGT (Relational Graph Transformer)
  • HGT (Heterogeneous Graph Transformer)
  • NBFNet (Neural Bellman-Ford Network)

Key Features:

  • Heterogeneous knowledge graph integration
  • Advanced transformer-based link prediction
  • Multi-modal data integration (microbial, chemical, environmental)

Related Projects:

  • Depends on: kg-microbe (knowledge graph foundation), MicroMediaParam (chemical compound mappings)
  • Feeds into: Media formulation recommendations, MicroGrowLinkService (API deployment)
  • Works with: MicroGrowAgents (complementary multi-agent predictions)

MicroGrowAgents

Dedicated Page | GitHub Repository β€” private repository, public release planned | Python | BSD-3-Clause

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

Specialized Agents:

  • LiteratureAgent - Mining 245+ papers for cultivation protocols
  • AnalogyReasoningAgent - Cross-organism comparison and reasoning
  • GenomeFunctionAgent - Auxotrophy detection from 57 Bakta-annotated genomes (667K features)
  • MediaFormulationAgent - Schema-driven media recommendation with evidence-based ingredient suggestions

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), chemical similarity (208K+ embeddings)
  • Genome-guided design for organism-specific media formulation

Related Projects:

  • Depends on: kg-microbe (knowledge graph foundation), MicroMediaParam (chemical mappings), eggnogtable (genome annotations), MATE-LLM (literature extraction)
  • Feeds into: Media formulation recommendations, PFASCommunityAgents (consortium design)
  • Works with: MicroGrowLink (complementary prediction approach)

KOGUT Transformer

DOE CODE 175162 | doi:10.11578/dc.20260210.3 | Released 2025-12-18

KOGUT (Knowledge Oriented Graph Unified Transformer) adapts the Relational Graph Transformer (RelGT) architecture β€” originally designed for relational tables and multi-table databases β€” to heterogeneous biological knowledge graphs, for link prediction over kg-microbe. Its primary task is predicting which growth media support a given microbial taxon.

Training data:

  • Merged kg-microbe knowledge graph: 1,392,337 nodes and 2,960,472 edges
  • 24 Biolink relation types spanning taxonomic hierarchies, metabolic interactions, phenotype associations, and environmental relationships
  • Primary task: growth media suitability (biolink:occurs_in, ~50K edges); the model can predict any of the 24 relation types
  • Trained on NVIDIA A100 GPUs at NERSC Perlmutter

Adaptations beyond the original RelGT:

  • Multimodal node encoding from KG metadata β€” labels, categories, descriptions, and synonyms
  • Extended k-hop subgraph sampling (3-hop default) tuned for sparse biological networks
  • Biolink predicate preservation, with type-specific transformations for the 24 edge semantics
  • Inductive learning, enabling zero-shot prediction for novel and uncultured taxa from feature-based embeddings (temperature, oxygen requirement, gram stain, cell shape)

Reported performance on growth media prediction: MRR 0.9966, Precision@1 0.9932, Hit@10 1.0000.

Status: registered in DOE CODE; no public source repository yet.

Related Projects:

  • Depends on: kg-microbe (training graph)
  • Works with: MicroGrowLink and MicroGrowAgents (complementary media-prediction approaches); explainable rule mining (interpretable counterpart)

MicroMediaParam

GitHub Repository | Python

Comprehensive chemical compound knowledge graph mapping pipeline for microbial growth media analysis. Extracts chemical compounds from media compositions and maps them to knowledge graph entities with standardized chemical properties.

Features:

  • Processes 23,181 chemical entries from 1,807 microbial growth media
  • 78% ChEBI coverage (18,088 compounds mapped)
  • Multi-database mapping to ChEBI, PubChem, and CAS-RN identifiers
  • Intelligent hydrate parsing and molecular weight calculation
  • Solution expansion for DSMZ solution references
  • 99.99% chemical mapping accuracy

Related Projects:

  • Depends on: CultureMech (chemical entity extraction)
  • Feeds into: kg-microbe (standardized chemical data), MicroGrowAgents (chemical mappings), MicroGrowLink (knowledge graph integration)
  • Works with: assay-metadata (compound identification)

CultureMech

Dedicated Page | GitHub Repository | Web Interface | CC BY 4.0 data Β· BSD-3-Clause code

15,915 curated culture media recipes deduplicated into 6,325 canonical media, with chemical entity extraction and ontology grounding. Part of the X-Mech suite of autonomous knowledge factories.

β†’ See the dedicated CultureMech page for full documentation, use cases, and examples.

Related Projects:

  • Depends on: Text-based media composition data
  • Feeds into: MicroMediaParam (entity mapping), kg-microbe (chemical data integration), MediaIngredientMech (ingredient curation)
  • Works with: assay-metadata (standardized substrate processing)

Specialized Research Pipelines

CMM-AI: Lanthanide Bioprocessing Data Pipeline

Private repository β€” public release planned | Python

Automated data pipeline for lanthanide bioprocessing research, focusing on rare earth element-dependent biological processes in microorganisms. Integrates multiple biological databases to create comprehensive research datasets.

Scientific Focus:

  • XoxF methanol dehydrogenase systems (lanthanide-dependent enzymes)
  • Methylotrophic bacteria (Methylobacterium, Methylorubrum, Paracoccus)
  • Environmental metal cycling and biogeochemistry
  • Siderophore/lanthanophore transport mechanisms
  • PQQ-dependent enzyme complexes

Related Projects:

  • Depends on: kg-microbe (organism data), eggnogtable (functional annotations)
  • Feeds into: Specialized lanthanide bioprocessing research
  • Works with: MicroGrowAgents (media optimization for lanthanide-dependent organisms)

PFAS-AI: Machine Learning-Enabled PFAS Biodegradation Pipeline

GitHub Repository | Python

ML-enabled data pipeline for PFAS biodegradation research, focusing on identification and characterization of microorganisms capable of degrading per- and polyfluoroalkyl substances (PFAS).

Research Objectives:

  • ML-Powered Database - Semantically-aware database using KG-Microbe platform to identify putative PFAS biodegradation genes, pathways, taxa, and environments
  • Intelligent Consortia Design - Graph learning and LLMs to design optimized microbial consortia for PFAS remediation

Scientific Focus:

  • C-F bond cleavage mechanisms (dehalogenases and defluorinases)
  • Fluoride resistance systems
  • Hydrocarbon degradation pathways
  • Environmental context (AFFF-contaminated sites, groundwater, wastewater)

Related Projects:

  • Depends on: kg-microbe (organism and gene identification)
  • Feeds into: PFASCommunityAgents (consortium design)
  • Works with: eggnogtable (functional gene annotations)

PFASCommunityAgents

Private repository β€” public release planned | Python

Multi-agent system for designing optimized microbial consortia for PFAS biodegradation. Uses AI-powered reasoning to compose consortia with complementary metabolic capabilities and syntrophic relationships.

Key Features:

  • Consortium composition optimization
  • Syntrophic relationship prediction
  • Environmental context-aware design
  • Multi-species compatibility assessment

Related Projects:

  • Depends on: PFAS-AI (candidate organism database), MicroGrowAgents (agent architecture), kg-microbe (organism relationships)
  • Feeds into: PFAS remediation research and consortium cultivation
  • Works with: MicroGrowAgents (media design for consortia)

Data Processing & Analysis

assay-metadata: BacDive API Assay Metadata Extractor

GitHub Repository | Python

Extracts API assay metadata from BacDive JSON data with comprehensive identifier mappings to CHEBI, EC, RHEA, and PubChem databases.

Capabilities:

  • Parses 99,392 bacterial strain records from BacDive
  • Extracts 17 unique API kit types (API zym, API 50CHac, etc.)
  • Maps substrate codes to CHEBI and PubChem identifiers
  • Maps enzyme EC numbers to RHEA reaction databases
  • Generates consolidated JSON metadata files

Related Projects:

  • Depends on: BacDive API data
  • Feeds into: kg-microbe (phenotypic assay data integration)
  • Works with: MicroMediaParam (compound identification), CultureMech (substrate processing)

eggnog_runner

GitHub Repository | Python

Automated pipeline for running EggNOG-mapper functional annotation at scale. Processes genome assemblies in batch to generate functional annotations for downstream analysis.

Features:

  • Batch genome processing with parallel execution
  • Automated EggNOG-mapper execution
  • Output standardization and organization
  • Error handling and retry logic

Related Projects:

  • Depends on: kg-microbe (genome data), EggNOG-mapper tool
  • Feeds into: eggnogtable (annotation post-processing)
  • Works with: MicroGrowAgents GenomeFunctionAgent (functional predictions)

eggnogtable

Private repository β€” public release planned | Python

Post-processing pipeline for EggNOG-mapper output into structured datasets. Extracts and organizes functional annotations including GO terms, EC numbers, and KEGG pathways.

Features:

  • GO term extraction and organization
  • EC number mapping
  • KEGG pathway assignment
  • Ontology term integration
  • Structured dataset generation

Related Projects:

  • Depends on: eggnog_runner (annotation output)
  • Feeds into: MicroGrowAgents GenomeFunctionAgent (auxotrophy detection), kg-microbe (functional annotations), CMM-AI (enzyme identification)
  • Works with: assay-metadata (enzyme EC number mapping)

microbe-rules: Machine Learning Models for Microbial Data

GitHub Repository | Python

Research code repository containing machine learning models and analysis pipelines for binary classification and comparative modeling of microbial datasets.

Features:

  • Binary classification models for microbial data
  • Model comparison and evaluation frameworks
  • Automated data preparation pipelines
  • Reproducible research workflows

Related Projects:

  • Depends on: kg-microbe (training data), various microbial datasets
  • Feeds into: Model optimization research
  • Works with: MicroGrowLink (model comparison), MicroGrowAgents (ML component evaluation)

AI Agent Systems

MATE-LLM

Private repository β€” public release planned | Python

LLM-powered system for extracting structured microbial information from scientific literature. Automates the extraction of cultivation protocols, growth conditions, and microbial annotations from research papers.

Key Features:

  • Entity extraction from scientific literature
  • Automated cultivation protocol annotation
  • Literature mining for growth conditions
  • Knowledge graph integration preparation
  • Structured data generation from unstructured text

Related Projects:

  • Depends on: Scientific literature corpus, LLM APIs
  • Feeds into: kg-microbe (literature-derived data), MicroGrowAgents LiteratureAgent (cultivation protocols)
  • Works with: METPO ontology (standardized terminology)

Web Services & APIs

MicroGrowLinkService

GitHub Repository | Python | REST API | BSD-3-Clause

RESTful API service wrapper for MicroGrowLink prediction models. Provides HTTP endpoints for programmatic access to growth media predictions and enables integration with laboratory information management systems (LIMS).

Key Features:

  • HTTP API endpoints for predictions
  • Model serving infrastructure
  • Batch prediction support
  • LIMS integration capabilities
  • Production deployment configuration

Related Projects:

  • Depends on: MicroGrowLink (prediction models), kg-microbe (knowledge graph)
  • Feeds into: External applications, LIMS integrations, web interfaces
  • Works with: MicroGrowAgents (complementary API services)

πŸ”¬ Advanced Research Tools

neurosymbolreason - Neurosymbolic Analogy Reasoning

GitHub Repository | Python

Neurosymbolic analogy reasoning on microbial knowledge graph embeddings to analyze relationships between microbial taxa and their physical growth preferences. Combines neural network embeddings with symbolic reasoning for cross-organism inference.

Key Features:

  • Knowledge graph embedding analysis
  • Analogy-based reasoning for growth preferences
  • Taxonomic relationship exploration
  • Novel organism growth condition inference

Related Projects:

  • Depends on: kg-microbe (knowledge graph embeddings), taxonomic data
  • Feeds into: MicroGrowAgents (AnalogyReasoningAgent), growth prediction pipelines
  • Works with: MicroGrowLink (complementary prediction approach)

auto-term-catalog - Automated Term Extraction

GitHub Repository | Python

Code for extracting AUTO terms from ontoGPT output. Processes ontology-based text mining results to create curated term catalogs for microbial cultivation research.

Key Features:

  • OntoGPT output processing
  • Automated term extraction and cataloging
  • Integration with METPO ontology
  • Standardized term generation

Related Projects:

  • Depends on: OntoGPT output, METPO ontology
  • Feeds into: kg-microbe (ontology terms), MATE-LLM (standardized vocabulary)
  • Works with: Literature mining pipelines

πŸ›  Developer Resources

culturebot-skills - Claude Code Skills

GitHub Repository | Skills/Configuration

Claude Code skills for CultureBot/KG-Microbe projects. Custom skills and workflows for AI-assisted development within the CultureBotAI ecosystem.

What it provides:

  • Pre-configured Claude Code skills for common tasks
  • Project-specific development workflows
  • Integration helpers for CultureBotAI tools
  • Best practices and code patterns

Use Cases:

  • Automated code generation for kg-microbe integrations
  • Data pipeline development assistance
  • Documentation generation
  • Testing and validation workflows

Getting Started:

# Install Claude Code skills
git clone https://github.com/CultureBotAI/culturebot-skills.git
# Follow setup instructions in repository README

πŸ“Š Datasets

Curated Cultivation Database

Curated collection of cultivation protocols for diverse microorganisms based on reference sources and literature.

Contents:

  • Growth media compositions
  • Environmental conditions (temperature, pH, atmosphere)
  • Cultivation methods and protocols
  • Literature references

Environmental Metadata Collection

Comprehensive dataset linking microorganisms to their natural habitats and environmental conditions.

🌐 Related Organizations & Resources

Academic & Research Institutions

  • ABPDU - Advanced Biofuels and Bioproducts Process Development Unit
  • BacDive - Bacterial Diversity Metadatabase
  • Cultivarium - Global microbial cultivation platform
  • JBEI - Joint BioEnergy Institute
  • JGI GOLD - Genomes Online Database
  • KBase - Systems Biology Knowledgebase
  • NMDC - National Microbiome Data Collaborative
  • Palsson Lab - UC San Diego Systems Biology Research Group

Commercial Organizations

  • Biolog - Microbial identification and characterization systems
  • Isolation Bio - Microbial isolation and cultivation technology

πŸ“š Documentation & Tutorials

API Documentation

Comprehensive documentation for programmatic access to kg-microbe and related tools:

  • Neo4j graph database interface
  • Python SDK usage examples
  • Data schema specifications

Tutorials

Coming soon!

Example Notebooks

Jupyter notebooks demonstrating practical applications:

  • Growth condition prediction workflows
  • Literature mining pipelines
  • Data visualization examples

πŸ”— Data Access & APIs

Direct Downloads

  • Knowledge Graph Dumps - Complete RDF/TTL files
  • Processed Datasets - CSV/JSON formatted data tables
  • Ontology Files - OWL/RDF ontology definitions

API Endpoints

Coming soon

Query Interfaces

Coming soon

πŸ“¦ Software Packages

Python Packages

Coming soon

🀝 Community & Collaboration

Contributing

We welcome contributions from the research community:

  • Data contributions - Share cultivation protocols and growth data
  • Software development - Contribute to open source tools
  • Literature curation - Help extract cultivation data from papers
  • Validation - Test predictions against experimental results

Discussion Forums

  • GitHub Discussions - Technical questions and feature requests
  • Slack Community - Real-time collaboration and support
  • Monthly Webinars - Updates and community presentations

Citation

If you use kg-microbe or other CultureBotAI resources in your research, please cite:

Santangelo, B.E., Hegde, H., Caufield, J.H., Reese, J., Kliegr, T., Hunter, L.E., 
Lozupone, C.A., Mungall, C.J., Joachimiak, M.P. (2026). KG-Microbe - Building 
Modular and Scalable Knowledge Graphs for Microbiome and Microbial Sciences. 
GigaScience, giag077. https://doi.org/10.1093/gigascience/giag077

Support & Contact

For technical support, collaboration inquiries, or questions about our resources:

Bibliography

  1. Santangelo BE, Hegde H, Caufield JH, Reese J, Kliegr T, Hunter LE, Lozupone CA, Mungall CJ, Joachimiak MP. KG-Microbe β€” Building Modular and Scalable Knowledge Graphs for Microbiome and Microbial Sciences. GigaScience. 2026;giag077. doi:10.1093/gigascience/giag077
  2. MΓ‘Ε‘a 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 Β· free full text
  3. 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
  4. Joachimiak MP. Knowledge Oriented Graph Unified Transformer (KOGUT) v0.1 [software]. DOE CODE; 2025. doi:10.11578/dc.20260210.3 Β· DOE CODE 175162
  5. Joachimiak MP, Santangelo BE, Hegde H, Caufield JH, Reese J, Kliegr T, Hunter LE, Lozupone CA, Mungall CJ. kg-microbe: modular knowledge graph for microbiome and microbial sciences [software]. github.com/Knowledge-Graph-Hub/kg-microbe
  6. METPO: Microbial Ecophysiological Trait and Phenotype Ontology. BioPortal Β· GitHub
  7. Joachimiak MP. β€œRuleML/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 β†’