MicroGrowAgents

Agent-based system for AI-driven microbial cultivation and growth media design

Part of the CultureBotAI initiative led by Dr. Marcin Joachimiak at Lawrence Berkeley National Laboratory.

Table of Contents

• Overview
• Key Achievements
• Key Features
• Agents & Skills
  • Core Agents (28)
  • Skills (50)
• Cofactor Analysis Data Sources
• Experimental Analysis & Optimization
  • Dual-Mode Analysis
  • Response Surface Modeling
  • Evidence-Based Interpretation
• Data Integrity & Provenance
  • Input Data Checksums
  • Artifact Cleanup
  • Audit Compliance
  • Citation Coverage
• Installation
• Quick Start
• Core Capabilities
• Advanced Usage
• Chemistry Modules
• Repository Structure
• Development
• Tools, APIs & Datasets
• Contributing
• Citation
• Contact

Overview

MicroGrowAgents bridges the microbial cultivation gap through AI-powered multi-agent systems that integrate knowledge graphs, machine learning, and experimental automation. The platform combines specialized agents (LiteratureAgent, AnalogyReasoningAgent, GenomeFunctionAgent, MediaFormulationAgent) operating on KG-Microbe (864,000+ validated species) to design optimized growth media for previously uncultured microorganisms.

📚 Documentation Quick Links:

• docs/STATUS.md - Current project state (start here)
• docs/AGENTS_SKILLS_TOOLS.md - Complete reference for agents, skills, and tools
• docs/OPTIMIZATION_GUIDE.md - Complete guide to data-driven v14 design
• docs/AUDIT_REPORT_BBOP_SKILLS.md - Audit compliance report (78% passing)
• CLAUDE.md - Guidance for Claude Code development

Key Achievements

• 🤖 MP_plus v10: Schema-driven media recommendation system with 15 evidence-based ingredient suggestions for Methylorubrum extorquens AM1 under lanthanide depletion stress
• 🧬 Genome-Guided Design: 57 Bakta-annotated genomes (667K features) for auxotrophy detection and organism-specific media formulation
• 📚 Knowledge Integration: 864,363 validated species across bacteria, archaea, fungi, and protozoa (GTDB + LPSN + NCBI)
• 🔬 Multi-Modal Reasoning: Literature mining (245+ papers), metabolic modeling (FBA/gap-filling), chemical similarity (208K+ embeddings), and experimental design
• ✅ Validated Outputs: 100% precision in organism extraction, complete toxicity transparency, schema-compliant output generation
• 🔒 Data Integrity: SHA256 checksums for all input data with cryptographic reproducibility tracking
• 📋 Audit Compliance: 78% compliance (7/9 PASS) against bbop-skills criteria for local-first agentic systems
• 📊 Citation Coverage: 90.5% (143/158 DOIs) with automated PDF retrieval and validation

Key Features

• 🧪 Media Concentration Predictions: Predict concentration ranges for media ingredients using ML-based regression
• 🔬 Advanced Chemistry Calculations:
  • Osmotic Properties: Osmolarity, osmolality, water activity, growth categories
  • Redox Properties: Eh (redox potential), pE, electron balance, redox state classification
  • Nutrient Ratios: C:N:P ratios, Redfield deviation, limiting nutrient identification, trace metal analysis
  • Thermodynamic Properties: Gibbs free energy calculations (via eQuilibrator API)
• 📊 Sensitivity Analysis: Sweep ingredient concentrations to determine pH and salinity effects
• 🔍 Media Comparison: Compare ingredient compositions across different media
• 🌐 External APIs: Integration with PubChem, ChEBI, and eQuilibrator for chemical data enrichment
• 📈 Visualization: Generate plots for osmotic properties, nutrient ratios, and sensitivity analysis
• 🤖 MP_plus Media Recommendation System: Multi-agent workflow generating complete media formulations with:
  • Literature-Based Discovery (Category 11): Organism-specific ingredient mining from 245+ papers
  • Analogy-Based Discovery (Category 7): Structural similarity search using 208K+ chemical embeddings
  • Genome-Guided Discovery (Categories 1-5): Metabolic modeling, auxotrophy detection, transporter analysis
  • Toxicity Flagging (Tier 2D): Transparent safety assessment (SAFE/CAUTION/WARNING)
  • Output Formats: YAML, TSV, CSV, JSON with complete provenance and validation
  • See data/designs/MP_plus/MP_plus_v10/ for example outputs
• 🧬 Genome Function Interpretation: Organism-specific media design using 57 Bakta-annotated genomes (667K features) with:
  • Auxotrophy Detection: Automatic identification of biosynthetic pathway gaps
  • Enzyme Analysis: EC number queries with wildcard support (1.1.. finds all CH-OH oxidoreductases)
  • Cofactor Requirements: Detection of essential cofactors that cannot be biosynthesized
  • Transporter Analysis: Concentration refinement based on nutrient uptake genes
  • See docs/GENOME_FUNCTION.md for Claude Code agent examples
• 📚 Sheet Query System: Query extended information sheets with:
  • 4 Query Types: Entity lookup, cross-reference, publication search, filtered queries
  • 3 Output Formats: Markdown tables, JSON, evidence-rich reports
  • Full-Text Search: Search within publication markdown files with excerpts
  • Cross-References: Automatic linking between entities and publications
  • See docs/SHEET_QUERY_SYSTEM.md for complete guide

Agents & Skills

MicroGrowAgents provides 28 specialized agents and 50 skills for microbial cultivation and media design.

Core Agents (28)

Knowledge & Reasoning:

• KGReasoningAgent - Query KG-Microbe knowledge graph (1.5M nodes, 5.1M edges)
• LiteratureAgent - Literature mining and evidence extraction
• AnalogyReasoningAgent - Chemical similarity search (208K+ embeddings)
• SheetQueryAgent - Query extended information sheets

Genome Analysis:

• GenomeFunctionAgent - Genome-guided media design (57 genomes, 667K features)
• LanthanideGenesAgent - Lanthanide-dependent gene analysis
• TransporterAgent - Nutrient transporter annotation and analysis

Media Design & Optimization:

• MediaFormulationAgent - Multi-source media recommendation
• GenMediaConcAgent - ML-based concentration prediction
• CofactorMediaAgent - Cofactor requirement analysis
• AlternateIngredientAgent - Alternative ingredient suggestions
• MediaRoleAgent - Ingredient metabolic role classification
• MaxProOptBlockAgent - MaxPro optimal blocking design generation
• ReconcileAgent - Experimental vs prediction reconciliation
• EnsembleOptimizationAgent - Response surface modeling and Bayesian optimization
• DesignRecommendationAgent - Interpret experimental results to recommend next design
• ExperimentalInterpretationAgent - Generate evidence-based biological interpretations with inline citations

Metabolic Modeling:

• MetabolicSourceAgent - Metabolic source identification
• GapMindAgent - GapMind pathway gap analysis integration
• GEMsemblerAgent - Genome-scale metabolic model reconstruction
• GrowthCodonAgent - Codon usage bias-based growth prediction
• MediaMatchAgent - MediaDive database integration

Chemistry & Properties:

• ChemistryAgent - Advanced chemistry calculations (osmotic, redox, nutrient ratios)
• MediapHCalculator - pH prediction and buffer design
• SensitivityAnalysisAgent - Parameter sweep and sensitivity analysis

Data Management:

• SQLAgent - Database queries and management
• IngredientCooccurrenceAgent - Ingredient co-occurrence analysis
• IngredientEffectsEnrichmentAgent - Ingredient effects enrichment
• CSVAllDOIsEnrichmentAgent - DOI-based literature enrichment
• PDFEvidenceExtractor - PDF evidence extraction
• EvidenceExtractionOrchestrator - Multi-source evidence orchestration

Skills (50)

Analysis Skills (19)

• analyze_cofactors - Cofactor requirements from genome annotations
• analyze_genome - Genome function interpretation (enzymes, auxotrophies, transporters)
• analyze_lanthanide_genes - Lanthanide-dependent gene analysis
• analyze_transporters - Transporter system analysis
• analyze_carbon_sources - Carbon source utilization analysis
• analyze_nitrogen_sources - Nitrogen source analysis
• analyze_phosphate_sources - Phosphate source analysis
• analyze_sulfur_sources - Sulfur source analysis
• analyze_sensitivity - pH and salinity sensitivity analysis
• analyze_cooccurrence - Ingredient co-occurrence patterns
• analyze_metabolic_requirements - Metabolic requirement analysis
• analyze_gaps - Metabolic pathway gap analysis
• analyze_limitations - Growth-limiting factor identification
• analyze_electron_balance - Electron donor/acceptor balance
• check_carbon_sources - Carbon source validation
• compare_auxotrophy_methods - Compare auxotrophy detection methods
• compare_gap_fba - Compare gap analysis with FBA
• annotate_transporters - Annotate transporter systems
• growth_prediction_dashboard - Interactive growth prediction dashboard
• interpret_experimental_results - Generate evidence-based biological interpretations

Prediction & Design Skills (12)

• predict_concentration - Predict ingredient concentration ranges
• predict_growth - Growth prediction from media composition
• predict_growth_cub - Codon usage bias-based growth prediction
• predict_growth_hybrid - Hybrid growth prediction (multiple methods)
• predict_transport_requirements - Predict transport requirements
• recommend_media - Media formulation recommendation
• recommend_media_quick - Quick media recommendation
• design_maxpro_optblock - MaxPro OptBlock experimental design
• optimize_growth_conditions - Ensemble optimization and Bayesian experiment design
• find_alternates - Find alternative ingredients
• classify_role - Classify ingredient metabolic roles
• reconstruct_model - Reconstruct genome-scale metabolic model

Query & Search Skills (5)

• query_knowledge_graph - Query KG-Microbe
• query_database - SQL database queries
• search_literature - Literature search and extraction
• search_mediadive - MediaDive database search
• sheet_query - Query extended information sheets

Chemistry & Validation Skills (5)

• calculate_chemistry - Calculate osmotic, redox, nutrient properties
• validate_media - Media formulation validation
• validate_formulation_comprehensive - Comprehensive formulation validation
• validate_ingredient - Ingredient validation and normalization
• export_results - Export results to multiple formats

Workflow Skills (6)

• recommend_media_workflow - Comprehensive media recommendation workflow
• recommend_media_comprehensive - Extended comprehensive workflow
• optimize_medium_workflow - Medium optimization workflow
• ingredient_report_workflow - Detailed ingredient analysis report
• initialize_database - Database initialization and validation
• export_results - Multi-format export utility

See src/microgrowagents/agents/ and src/microgrowagents/skills/ for complete documentation.

Cofactor Analysis Data Sources

The CofactorMediaAgent integrates 6 major biological databases and specialized literature:

Primary Databases

• ChEBI - Chemical identifiers for 44 cofactors (DOI: 10.1093/nar/gkv1031)
• KEGG - 30+ biosynthesis pathway definitions (DOI: 10.1093/nar/gkac963)
• BRENDA - EC-to-cofactor relationships (DOI: 10.1093/nar/gky1048)
• ExplorEnz - Enzyme Commission nomenclature (DOI: 10.1093/nar/gkn582)

Knowledge Graph Integration

• KG-Microbe (1.5M nodes, 5.1M edges) - Enzyme-substrate relationships and pathway context
• Queries via KGReasoningAgent for multi-source evidence integration

Reference Files

• src/microgrowagents/data/cofactor_hierarchy.yaml - 44 cofactors across 5 categories
• src/microgrowagents/data/ec_to_cofactor_map.yaml - 68 EC pattern mappings
• data/processed/ingredient_cofactor_mapping.csv - 13 MP medium cofactor providers

See docs/cofactor_data_sources.md for detailed methodology and citations.

Example: Cofactor Analysis for M. extorquens AM-1

Generate cofactor requirements table from Bakta genome annotations:

# Using Python API
uv run python -c "
from microgrowagents.agents import CofactorMediaAgent
from pathlib import Path

agent = CofactorMediaAgent(Path('data/processed/microgrow.duckdb'))
result = agent.run(
    query='Analyze cofactor requirements',
    organism='SAMN31331780',  # M. extorquens AM-1
    base_medium='MP'
)

# Save results
import pandas as pd
df = pd.DataFrame(result['data']['cofactor_table'])
df.to_csv('outputs/cofactor_analysis/cofactor_table_Methylorubrum_extorquens_AM1.csv')
"

Results for M. extorquens AM-1 (from 110 EC numbers):

• 15 cofactors identified
• 4 existing in MP medium: TPP, Biotin, Fe-S clusters, Mg
• 11 missing: PLP, THF, Coenzyme Q, NAD+, NADP+, ATP, CTP, GTP, UTP, CoA, SAM

Generated tables available at:

• CSV: outputs/cofactor_analysis/cofactor_table_Methylorubrum_extorquens_AM1.csv
• TSV: outputs/cofactor_analysis/cofactor_table_Methylorubrum_extorquens_AM1.tsv

Experimental Analysis & Optimization

MicroGrowAgents provides a comprehensive dual-pipeline for analyzing experimental growth data with both absolute and relative analysis modes, plus response surface modeling and Bayesian optimization.

Features

• 📊 Dual-Mode Analysis: Absolute (raw OD600) and Relative (vs baseline) analysis pipelines
• 🔬 Hierarchical Clustering: Identify groups of similar growth conditions (276 replicates, 6 clusters)
• 🗺️ Response Surface Modeling: Gaussian Process modeling with multi-objective optimization and Pareto frontiers
• 🤖 Ensemble Optimization: Gaussian Process, Polynomial, and Random Forest ensemble models
• 🎯 Bayesian Optimization: Adaptive experiment design with Expected Improvement acquisition
• 📈 Effect Analysis: ANOVA, main effects plots, Sobol sensitivity indices
• ✅ Schema-Driven Validation: Automatic validation of all analysis outputs with source data traceability
• 🏷️ Output Labeling: All outputs labeled with source experimental data ID for full provenance
• 🔍 Evidence-Based Interpretation: Automated biological interpretation with inline citations and bibliography

Quick Start: Analyze Experimental Data

Analyze experimental plate data with dual-mode analysis (absolute + relative):

# Run BOTH absolute and relative analyses (recommended)
just analyze-experimental data/experimental/plate_designs_v10_maxprooptblock_long__results

# Run only absolute analysis (raw OD600)
just analyze-experimental-absolute data/experimental/plate_designs_v10_maxprooptblock_long__results

# Run only relative analysis (fold-change vs control)
just analyze-experimental-relative data/experimental/plate_designs_v10_maxprooptblock_long__results

# Run clustering on results
just cluster-experimental outputs/plate_designs_v10_maxprooptblock_long__results_experimental_analysis_absolute/v10_maxprooptblock_long__results_replicate_statistics_absolute.tsv outputs/plate_designs_v10_maxprooptblock_long__results_experimental_analysis_clustering_absolute absolute

# Validate all outputs
just validate-experimental plate_designs_v10_maxprooptblock_long__results

Analysis Modes:

• Absolute Analysis: Raw OD600 measurements showing actual biomass achieved

  • Answers: "Which conditions grew best overall?"
  • Use for: Identifying highest-performing conditions, comparing to literature values

• Relative Analysis: Fold-change, difference, and percent change vs control baseline

  • Answers: "Which variations improved over baseline media?"
  • Use for: Identifying growth enhancements, normalizing across experiments

Pipeline Steps:

1. Statistical Processing → v10_maxprooptblock_long__results_replicate_statistics_{mode}.tsv
2. Exploratory Visualization → v10_maxprooptblock_long__results_growth_curves.pdf
3. Hierarchical Clustering → v10_maxprooptblock_long__results_clustered_heatmap_growth.pdf
4. Response Surface Modeling → response_surfaces/surface_3d_{measurement}_{mode}.pdf (optional)
5. Output Validation → All files verified with proper source data ID labeling

Output Directories:

• Absolute analysis: outputs/{source_data_id}_experimental_analysis_absolute/
• Relative analysis: outputs/{source_data_id}_experimental_analysis_relative/
• Clustering results: outputs/{source_data_id}_experimental_analysis_clustering_{mode}/

Response Surface Modeling

The experimental analysis pipeline includes optional response surface modeling using Gaussian Processes to understand ingredient-measurement relationships and multi-objective optimization:

# Response surfaces run automatically with analyze-experimental (enabled by default)
just analyze-experimental data/experimental/plate_designs_v13_latinhypercube_long__results

# Disable response surfaces for faster analysis
python scripts/run_dual_analysis.py data/experimental/plate_designs_v10_maxprooptblock_long__results --disable-response-surfaces

# Standalone response surface analysis
python scripts/analyze_response_surfaces.py \\
    outputs/plate_designs_v13_latinhypercube_long__results_experimental_analysis_absolute/ \\
    --mode absolute \\
    --measurements OD600 Nd_uM

Capabilities:

• 🗺️ 3D Surface Plots: Visualize ingredient-measurement relationships
• 🎯 Pareto Frontiers: Multi-objective optimization (e.g., maximize OD600 while minimizing Nd consumption)
• 🔮 Predictions: Predict measurements over entire design space
• 📊 Contour Maps: Identify optimal ingredient combinations

Use Cases:

• v13+ designs with variable Neodymium for lanthanide-dependent growth analysis
• Understanding growth-lanthanide relationships (MxaF vs XoxF-MDH pathways)
• Identifying optimal conditions for multiple objectives simultaneously

Measurement Types:

• OD600 (Optical Density): Bacterial biomass (higher = more growth)
  • Absolute mode: Raw OD600 values
  • Relative mode: Fold-change vs control baseline
• Nd_uM (Neodymium concentration): Lanthanide depletion marker
  • Values relative to baseline media WITH bacterial growth
  • Negative values: More Nd consumption than control (higher bacterial uptake)
  • Positive values: Less Nd consumption than control (lower bacterial uptake)
  • Used to distinguish lanthanide-dependent vs independent growth pathways

Outputs (per mode):

• response_surfaces/surface_predictions_{measurement}_{mode}.csv - Predictions over design space
• response_surfaces/surface_3d_{measurement}_{mode}.pdf/png - 3D surface plots
• response_surfaces/pareto_frontier_{mode}.csv - Pareto-optimal conditions (joint analysis)
• response_surfaces/pareto_frontier_{mode}.pdf/png - Pareto frontier visualization
• response_surfaces/optimization_report_{mode}.txt - Model parameters and best conditions

Optimization Workflow

Build response surface models and suggest next experiments using ensemble modeling:

# Using Python skill with source data ID (recommended)
uv run python -m microgrowagents.skills.simple.optimize_growth_conditions \
    --data outputs/experimental_analysis \
    --source-data-id plate_designs_v10_maxprooptblock_long__results \
    --output-dir outputs/optimization \
    --strategy hybrid \
    --n-suggestions 69

# Or via direct file path
uv run python -m microgrowagents.skills.simple.optimize_growth_conditions \
    --data outputs/experimental_analysis/v10_maxprooptblock_long__results_replicate_statistics.tsv \
    --output-dir outputs/optimization \
    --strategy hybrid

What it does:

1. Trains ensemble models (Gaussian Process + Polynomial + Random Forest)
2. Analyzes ingredient effects and interactions
3. Uses Bayesian optimization to suggest next experiments
4. Generates v12 design files compatible with pipetting infrastructure

Optimization Strategies:

• Bayesian Optimization: Expected Improvement acquisition (exploitation)
• Local Search: Perturbation around best observed conditions
• Uncertainty Sampling: Explore high-uncertainty regions (exploration)
• Hybrid: 70% local search + 15% uncertainty + 15% space-filling

Key Results: M. extorquens AM-1 Lanthanide Study

v10 Design - 69 conditions tested (4 replicates each, 3 timepoints):

Top Performer: MPOB_040

• Max OD600: 0.95 (highest overall)
• Strategy: Pure C1 methylotrophy (67.9 mM methanol, low succinate)
• Challenge: 98% crash at 48h due to methanol depletion

Most Stable: MPOB_053

• Max OD600: 0.66 (sustained growth)
• Strategy: Mixed C1+C2 metabolism (19.9 mM methanol, 58.7 mM succinate)
• Result: Stable across all timepoints

Key Finding: 40-60 mM succinate provides metabolic backup when methanol depletes, preventing culture crash while maintaining high peak growth.

v13 Design - Lanthanide-Dependent Growth Pathways:

• Variable Neodymium (0-5 µM) to test MxaF vs XoxF-MDH pathways
• Response surface modeling identifies Pareto-optimal conditions
• High OD600 at low Nd → lanthanide-independent pathway (MxaF-MDH)
• High OD600 at high Nd → lanthanide-dependent pathway (XoxF-MDH)
• Multi-objective optimization balances growth AND Nd utilization

See outputs/optimization/MPOB_040_CRASH_ANALYSIS.md for detailed v10 analysis.

Schema & Validation

All analysis outputs conform to validation standards with automatic source data ID labeling:

• Validator: src/microgrowagents/utils/analysis_output_validator.py
• Documentation: docs/EXPERIMENTAL_ANALYSIS_PIPELINE.md

Source Data Traceability:

The system automatically generates output prefixes from source data directories:

Input Directory:  plate_designs_v10_maxprooptblock_long__results
       ↓
Output Prefix:    v10_maxprooptblock_long__results_
       ↓
Output Files:     v10_maxprooptblock_long__results_replicate_statistics.tsv
                  v10_maxprooptblock_long__results_growth_curves.pdf
                  v10_maxprooptblock_long__results_clustered_heatmap_growth.pdf
                  v10_maxprooptblock_long__results_cluster_assignments_growth.csv

Prefix Generation:

• Removes plate_designs_ from source directory name
• Adds trailing underscore
• Applies to all outputs: statistical, visualization, and clustering files

Every output file is:

• ✅ Labeled with source experimental data ID for full traceability
• ✅ Named consistently across all analysis types
• ✅ Validated for existence and proper formatting
• ✅ Documented with file counts and metadata

Example: Complete Output Set

For source data plate_designs_v10_maxprooptblock_long__results, the pipeline generates:

Statistical Analysis (outputs/plate_designs_v10_maxprooptblock_long__results_experimental_analysis_{mode}/):

• v10_maxprooptblock_long__results_processed_data_raw.tsv
• v10_maxprooptblock_long__results_processed_data_{mode}.tsv (absolute or relative)
• v10_maxprooptblock_long__results_replicate_statistics_{mode}.tsv
• v10_maxprooptblock_long__results_control_statistics.tsv

Visualization (outputs/plate_designs_v10_maxprooptblock_long__results_experimental_analysis_{mode}/):

• v10_maxprooptblock_long__results_growth_curves.pdf/.png
• v10_maxprooptblock_long__results_dose_response_curves.pdf/.png
• v10_maxprooptblock_long__results_heatmaps.pdf/.png
• v10_maxprooptblock_long__results_pca_ingredient_space.pdf/.png
• v10_maxprooptblock_long__results_pca_measurement_space.pdf/.png
• v10_maxprooptblock_long__results_replicate_variability.pdf/.png
• v10_maxprooptblock_long__results_summary_statistics.pdf/.png

Clustering (outputs/plate_designs_v10_maxprooptblock_long__results_experimental_analysis_clustering_{mode}/):

• v10_maxprooptblock_long__results_clustered_heatmap_growth.pdf/.png
• v10_maxprooptblock_long__results_cluster_assignments_growth.csv
• v10_maxprooptblock_long__results_cluster_descriptions_growth.txt
• v10_maxprooptblock_long__results_cluster_summary_growth.pdf

Response Surfaces (optional, outputs/plate_designs_v10_maxprooptblock_long__results_experimental_analysis_{mode}/response_surfaces/):

• surface_predictions_{measurement}_{mode}.csv
• surface_3d_{measurement}_{mode}.pdf/.png
• pareto_frontier_{mode}.csv (multi-objective optimization)
• pareto_frontier_{mode}.pdf/.png
• optimization_report_{mode}.txt

Evidence-Based Interpretation

Generate publication-ready biological interpretations with inline citations and bibliography:

from microgrowagents.agents.analysis import ExperimentalInterpretationAgent

# Initialize agent with version identifier
agent = ExperimentalInterpretationAgent(source_version="v10")

# Run interpretation workflow
result = agent.run()

What it generates:

1. INTERPRETATION_REPORT.md - Clean biological interpretation

   • Executive summary with key findings
   • Factor-by-factor analysis (phosphate, nitrogen, carbon sources)
   • Metabolic insights (carbon utilization, nutrient stoichiometry)
   • Evidence-based hypotheses with testable predictions
   • Recommendations for next design iteration
   • Optimal media formulation based on results

2. INTERPRETATION_EVIDENCE.md - Evidence companion file

   • Data Evidence (E1-E#) with specific file references:
     • E1: Control statistics from v10_..._control_statistics.tsv
     • E2: Top 10 conditions from v10_..._replicate_statistics.tsv
     • E3: Clustering patterns from v10_..._cluster_descriptions_growth.txt
     • E4: Boundary effects from DesignRecommendationAgent analysis
   • Literature Evidence (L1-L#) with DOIs:
     • L1: M. extorquens metabolism (Chistoserdova et al. 2003)
     • L2: PQQ-dependent MDH (Anthony & Williams 2003)
     • L3: Rare earth elements (Pol et al. 2014)
   • Each evidence includes: source file, full path, section, data snippet

3. INTERPRETATION_REPORT_evidence.md - Citation-based report

   • Same content as main report but with inline citations [E1], [E2], [L1], [L2]
   • Complete bibliography with file references and data snippets
   • Publication-ready format

4. interpretation_metadata.json - Execution metadata

   • Timestamp, directories used, summary statistics

Example output:

================================================================================
ExperimentalInterpretationAgent - Evidence-Based Interpretation
================================================================================

Step 1: Locating analysis directories...
  ✓ Analysis directory: outputs/plate_designs_v10_.../
  ✓ Clustering directory: outputs/plate_designs_v10_..._clustering/

Step 2: Validating data files...
  ✓ Required data files present

Step 3: Generating interpretation reports...
  - Analyzing experimental data...
  - Extracting evidence snippets...
  - Generating biological interpretation...
  - Creating citation-based report...

Step 4: Interpretation complete!

Summary:
  Conditions analyzed: 10
  Clusters identified: 6
  Boundary effects: 3
  Evidence snippets: 4
  Literature references: 3

Key Features:

• 📚 Complete traceability: Every claim cites specific data files and sections
• 🔬 Biological insights: Factor-by-factor interpretation with metabolic context
• 📊 Data snippets: Actual values from analysis files included in bibliography
• 📖 Literature support: DOI-linked references with key findings
• ✅ Publication-ready: Three formats (clean, evidence, citation-based)

See: docs/EXPERIMENTAL_INTERPRETATION_AGENT.md for complete documentation

Data Integrity & Provenance

MicroGrowAgents implements comprehensive data integrity and provenance tracking for reproducibility:

Input Data Checksums

All input data files are protected with SHA256 checksums for cryptographic reproducibility:

# Verify input data integrity
just verify-data-integrity

# Generate checksums for new data
python scripts/generate_checksums.py data/raw/

Checksums stored in:

• Global checksums: data/checksums.txt
• Per-analysis checksums: outputs/*/input_data_checksums.json

Automatic tracking:

• Every analysis records checksums of input files
• Verification detects any data modifications or corruption
• Complies with bbop-skills Criterion 4 (cryptographic reproducibility)

Artifact Cleanup Policy

Three-tier retention model for efficient storage management:

Archival (Keep indefinitely):

• Published experimental designs (v10, v13, etc.)
• Validated analysis results with interpretations
• Response surface models

Temporary (30 days):

• Experimental analysis outputs
• Clustering results
• Intermediate optimization runs

Ephemeral (7 days):

• Test outputs
• Debugging artifacts
• Temporary visualizations

Cleanup Commands:

# Archive old outputs (moves to archive/ directory)
just archive-outputs

# Clean old outputs (>30 days)
just clean-old-outputs

# Clean ephemeral artifacts (>7 days)
just clean-ephemeral

Storage Impact:

• Steady-state: ~185MB (with cleanup)
• Unmanaged: ~4GB/year (96% reduction)

See: docs/ARTIFACT_CLEANUP_POLICY.md for complete retention policies

Audit Compliance

Overall Compliance: 78% (7/9 PASS) against bbop-skills criteria:

✅ PASS (7 criteria):

• Provenance tracking (.claude/provenance/)
• Model tracking (explicit model IDs in all outputs)
• Reasoning/code separation (markdown interpretations + code artifacts)
• Validation (LinkML schemas, output validators)
• Error-correction (DOI validation + corrections)
• RAG (KG-Microbe, literature corpus, genome annotations)
• Artifact cleanup (automated retention policies)

⚠️ PARTIAL (1 criterion):

• Documentation/automation (needs enhancement)

❌ FAIL (1 criterion):

• MCP integration (not yet adopted, under consideration)

See: docs/AUDIT_REPORT_BBOP_SKILLS.md for complete audit findings

Citation Coverage

DOI Validation: 90.5% (143/158 DOIs) with evidence

• PDFs: 92 (58.2%)
• Abstracts: 44 (27.8%)
• Missing: 15 (9.5%)

Automated Workflows:

# Validate DOIs
uv run python scripts/doi_validation/validate_failed_dois.py

# Apply corrections
uv run python scripts/doi_corrections/apply_doi_corrections.py

# Download PDFs
uv run python scripts/pdf_downloads/download_all_pdfs_automated.py

See: notes/DOI_CORRECTIONS_FINAL_UPDATED.md for correction history

Installation

Prerequisites

• Python 3.10 or higher
• uv package manager

Quick Install

# Clone the repository
git clone https://github.com/CultureBotAI/MicroGrowAgents.git
cd MicroGrowAgents

# Install dependencies using uv
uv sync

# Verify installation
uv run python run.py --help

Quick Start

Generate Media Concentrations

Predict concentration ranges for a specific medium:

# Get MP medium concentrations
uv run python run.py gen-media-conc "MP medium"

# Get concentrations for custom ingredients
uv run python run.py gen-media-conc "glucose,NaCl,KH2PO4" --mode ingredients

# Export to JSON
uv run python run.py gen-media-conc "MP medium" --format json --output mp_medium.json

Sensitivity Analysis

Analyze how ingredient concentration variations affect pH and salinity:

# Basic sensitivity analysis
uv run python run.py sensitivity "MP medium"

# With osmotic property calculations
uv run python run.py sensitivity "MP medium" --calculate-osmotic

# With all advanced properties
uv run python run.py sensitivity "MP medium" \
    --calculate-osmotic \
    --calculate-redox \
    --calculate-nutrients \
    --plot

# Custom parameters
uv run python run.py sensitivity "glucose,NH4Cl,KH2PO4" \
    --calculate-redox \
    --ph 6.5 \
    --temperature 37

Advanced Chemistry Analysis

Calculate osmotic properties for a medium:

from microgrowagents.chemistry.osmotic_properties import (
    calculate_osmolarity,
    calculate_water_activity
)

ingredients = [
    {"name": "NaCl", "concentration": 150.0, "molecular_weight": 58.44, "formula": "NaCl"},
    {"name": "KCl", "concentration": 5.0, "molecular_weight": 74.55, "formula": "KCl"}
]

# Calculate osmolarity
osm_result = calculate_osmolarity(ingredients, temperature=25.0)
print(f"Osmolarity: {osm_result['osmolarity']:.1f} mOsm/L")

# Calculate water activity
aw_result = calculate_water_activity(ingredients, temperature=25.0)
print(f"Water Activity: {aw_result['water_activity']:.4f}")
print(f"Growth Category: {aw_result['growth_category']}")

Core Capabilities

1. Media Concentration Generation (gen-media-conc)

Predicts LOW, DEFAULT, and HIGH concentration ranges for media ingredients:

# Query by medium name
uv run python run.py gen-media-conc "MP medium"

# Query by ingredient list
uv run python run.py gen-media-conc "PIPES,NaCl,glucose" --mode ingredients

# With chemical data enrichment
uv run python run.py gen-media-conc "MP medium" --enrich pubchem

Output includes:

• Predicted concentration ranges (mM)
• Molecular weights
• Chemical formulas
• Confidence scores

2. Sensitivity Analysis (sensitivity)

Performs parameter sweep analysis by varying each ingredient between LOW and HIGH concentrations:

# Basic analysis (pH and salinity)
uv run python run.py sensitivity "MP medium"

# With advanced chemistry properties
uv run python run.py sensitivity "MP medium" --calculate-osmotic --calculate-nutrients

# Export results
uv run python run.py sensitivity "MP medium" --format json --output results.json

# Generate visualization
uv run python run.py sensitivity "MP medium" --plot --plot-output analysis.png

Calculates:

• pH changes
• Salinity (TDS and NaCl-equivalent)
• Ionic strength
• Optional: Osmotic properties, redox potential, nutrient ratios

3. Advanced Chemistry Properties

Osmotic Properties

Calculate osmolarity, osmolality, and water activity:

uv run python run.py sensitivity "MP medium" --calculate-osmotic

Provides:

• Osmolarity (mOsm/L)
• Osmolality (mOsm/kg)
• Water activity (aw)
• Growth category classification:
  • most_bacteria (aw > 0.98)
  • halotolerant (0.90 < aw ≤ 0.98)
  • halophiles (aw ≤ 0.90)
• Van't Hoff dissociation factors

Example output:

{
  "osmotic_properties": {
    "osmolarity": 342.5,
    "osmolality": 339.8,
    "water_activity": 0.9938,
    "growth_category": "most_bacteria",
    "confidence": {"osmolarity": 0.85, "water_activity": 0.78}
  }
}

Redox Properties

Calculate redox potential (Eh), pE, and electron balance:

uv run python run.py sensitivity "glucose,NH4Cl" --calculate-redox --ph 7.0

Calculates:

• Eh (redox potential in mV)
• pE (electron activity)
• Redox state classification (oxidizing, reducing, intermediate)
• Electron donor/acceptor balance
• Standard redox couples (O2/H2O, NO3-/NO2-, SO42-/H2S, etc.)

Uses Nernst equation:

Eh = E0' + (59.16/n) × log([oxidized]/[reduced])  at 25°C
pH correction: Eh = E0 - (59.16/n) × pH

Example output:

{
  "redox_properties": {
    "eh": 245.3,
    "pe": 4.15,
    "redox_state": "oxidizing",
    "electron_balance": {
      "total_donors": 240.0,
      "total_acceptors": 220.0,
      "balance": 8.3
    }
  }
}

Nutrient Ratios

Calculate C:N:P ratios and identify limiting nutrients:

uv run python run.py sensitivity "glucose,NH4Cl,KH2PO4" --calculate-nutrients

Analyzes:

• C:N:P molar ratios
• Limiting nutrient prediction
• Redfield ratio deviation (marine standard: 106:16:1)
• Trace metal ratios (Fe:P, Mn:P, Zn:P)
• Deficiencies and excesses

Limiting nutrient criteria:

• P-limited: C:P > 150 or N:P > 20
• N-limited: C:N > 20 or N:P < 10
• C-limited: C:N < 6.6
• Balanced: Near Redfield ratio

Example output:

{
  "nutrient_ratios": {
    "c_mol": 60.0,
    "n_mol": 9.0,
    "p_mol": 0.6,
    "c_n_ratio": 6.67,
    "c_p_ratio": 100.0,
    "n_p_ratio": 15.0,
    "limiting_nutrient": "balanced",
    "redfield_deviation": 3.2,
    "trace_metals": {
      "fe_p_ratio": 0.015,
      "deficiencies": ["Co", "Mo"],
      "excesses": []
    }
  }
}

4. Media Comparison

Compare ingredient compositions between two media:

uv run python run.py compare-media "MP medium" "LB medium"

Shows:

• Common ingredients
• Unique ingredients to each medium
• Concentration differences

5. Media Formulation Recommendation (recommend-media workflow)

Recommend new media formulations using AI-powered multi-agent orchestration:

from microgrowagents.skills.workflows import RecommendMediaWorkflow

# Initialize workflow
workflow = RecommendMediaWorkflow()

# Recommend organism-specific medium
result = workflow.run(
    query="Recommend medium for methanotrophic bacteria",
    organism="Methylococcus capsulatus",
    temperature=42.0,
    pH=6.8,
    carbon_source="methane",
    oxygen="aerobic",
    goals="defined,selective",
    output_format="markdown"
)
print(result)

Features:

• Multi-source Evidence Integration: Combines KG-Microbe, literature, and MP database
• Organism-Specific: Tailored to target organism metabolic requirements
• Complete Formulation: Ingredient list with concentrations, roles, and confidence scores
• Chemical Compatibility: Validates precipitation and antagonism risks
• Alternative Ingredients: Provides substitutes with rationales
• Comprehensive Rationale: Human-readable explanations for all decisions

Example Goals:

• minimal - Fewest ingredients, core nutrients only
• defined - All ingredients chemically defined, no undefined supplements
• complex - Rich nutrients, may include vitamins and cofactors
• cost_effective - Prioritizes inexpensive, common ingredients
• high_yield - Optimized for biomass/product formation
• selective - Includes selective agents or unusual nutrients

Output includes:

• Complete ingredient list with concentrations and ranges
• Predicted pH, ionic strength, and other properties
• Essential nutrient roles coverage
• Chemical compatibility notes
• Alternative ingredient suggestions
• Evidence from KG-Microbe, literature, and database
• Confidence scoring based on evidence quality

See .claude/skills/recommend-media.md for detailed documentation and examples.

6. Genome Function Interpretation

Organism-specific media design using Bakta-annotated genomes (57 genomes, 667,502 features):

Key Capabilities:

• Auxotrophy Detection: Automatically identify biosynthetic pathway gaps
• Enzyme Queries: EC number searches with wildcard support (e.g., 1.1.*.*)
• Cofactor Analysis: Determine essential cofactors that cannot be biosynthesized
• Transporter Analysis: Find nutrient uptake genes for concentration refinement

CLI Examples:

# Find oxidoreductase enzymes
from microgrowagents.agents.kg_reasoning_agent import KGReasoningAgent
from pathlib import Path

agent = KGReasoningAgent(Path('data/processed/microgrow.duckdb'))
result = agent.run('genome_enzymes SAMN00114986 1.1.*')
print(f"Found {result['data']['count']} enzymes")

# Detect auxotrophies
from microgrowagents.agents.genome_function_agent import GenomeFunctionAgent

agent = GenomeFunctionAgent(Path('data/processed/microgrow.duckdb'))
result = agent.detect_auxotrophies(query='detect auxotrophies', organism='SAMN00114986')
print(f"Detected {result['data']['summary']['auxotrophies_detected']} auxotrophies")

Claude Code Agent Examples:

See docs/GENOME_FUNCTION.md for detailed examples including:

• Analyzing organism metabolic capabilities
• Comparing metabolic profiles of different organisms
• Designing organism-specific defined media
• Auxotrophy-guided media optimization
• Metabolic engineering context analysis

Automatic Integration:

Genome analysis is automatically integrated into:

• MediaFormulationAgent: Adds nutrients for detected auxotrophies
• GenMediaConcAgent: Refines concentrations based on transporter presence/affinity
• KGReasoningAgent: Adds genome_enzymes, genome_auxotrophies, genome_transporters queries

7. Claude Code Skills

The MicroGrowAgents skills framework provides 18 Claude Code skills:

Simple Skills (12 skills)

Cofactor Analysis Skill:

from microgrowagents.skills.simple import AnalyzeCofactorsSkill

skill = AnalyzeCofactorsSkill()
result = skill.run(
    organism="SAMN31331780",  # M. extorquens AM-1
    base_medium="MP",
    output_format="markdown"
)
print(result)

Genome Analysis Skill:

from microgrowagents.skills.simple import AnalyzeGenomeSkill

skill = AnalyzeGenomeSkill()
result = skill.run(
    query="Find all methanol dehydrogenases",
    organism="SAMN31331780",
    analysis_type="enzymes",
    ec_pattern="1.1.2.*",
    output_format="markdown"
)
print(result)

Knowledge Graph Query Skill:

from microgrowagents.skills.simple import QueryKnowledgeGraphSkill

skill = QueryKnowledgeGraphSkill()
result = skill.run(
    query="Find media for Methylococcus capsulatus",
    query_type="organism_media",
    output_format="markdown"
)
print(result)

Other Simple Skills:

• PredictConcentrationSkill - Predict ingredient concentrations
• FindAlternatesSkill - Find alternative ingredients
• AnalyzeSensitivitySkill - Sensitivity analysis for pH/salinity
• ClassifyRoleSkill - Classify ingredient metabolic roles
• SearchLiteratureSkill - Search scientific literature
• QueryDatabaseSkill - SQL queries on MP medium database
• CalculateChemistrySkill - Calculate osmotic/redox/nutrient properties
• AnnotateTransportersSkill - Annotate transporter systems in genomes
• PredictTransportRequirementsSkill - Predict transport requirements for medium ingredients

Workflow Skills (3 workflows)

• RecommendMediaWorkflow - Comprehensive media formulation recommendation
• OptimizeMediumWorkflow - Medium optimization for specific goals
• IngredientReportWorkflow - Detailed ingredient analysis reports

Utility Skills (3 utilities)

• InitializeDatabaseSkill - Database initialization and validation
• ExportResultsSkill - Export results to JSON/CSV/Excel
• ValidateIngredientSkill - Ingredient validation and normalization

See src/microgrowagents/skills/ for complete skill documentation.

8. Integration Scripts

Standalone integration scripts for specific analyses:

# MP medium with osmotic properties
uv run python scripts/analyze_mp_medium_osmotic.py --plot --output-json results.json

# Generate visualization plots
uv run python scripts/analyze_mp_medium_osmotic.py --plot --plot-output mp_osmotic.png

Advanced Usage

Combining Multiple Property Calculations

Calculate all advanced properties simultaneously:

uv run python run.py sensitivity "MP medium" \
    --calculate-osmotic \
    --calculate-redox \
    --calculate-nutrients \
    --ph 7.0 \
    --temperature 30 \
    --format json \
    --output complete_analysis.json

Pipeline Mode

Use gen-media-conc output as input to sensitivity:

# Step 1: Generate concentration predictions
uv run python run.py gen-media-conc "MP medium" --format json > predictions.json

# Step 2: Run sensitivity analysis on predictions
uv run python run.py sensitivity --input-file predictions.json --calculate-osmotic

Python API

Use MicroGrowAgents programmatically:

from microgrowagents.agents.sensitivity_analysis_agent import SensitivityAnalysisAgent

# Initialize agent
agent = SensitivityAnalysisAgent(db_path="data/microgrowdb.db")

# Run analysis with advanced properties
result = agent.run(
    query="MP medium",
    mode="medium",
    calculate_osmotic=True,
    calculate_redox=True,
    calculate_nutrients=True,
    temperature=37.0
)

# Access results
baseline = result["baseline"]
print(f"pH: {baseline['ph']}")
print(f"Osmolarity: {baseline['osmotic_properties']['osmolarity']} mOsm/L")
print(f"Limiting nutrient: {baseline['nutrient_ratios']['limiting_nutrient']}")

Chemistry Modules

Osmotic Properties

Module: microgrowagents.chemistry.osmotic_properties

Functions:

• calculate_osmolarity(ingredients, temperature=25.0) - Calculate osmolarity and osmolality
• calculate_water_activity(ingredients, temperature=25.0, method="raoult") - Calculate water activity
• estimate_van_hoff_factor(formula, charge, name) - Estimate dissociation factor

Methods:

• Raoult's law (dilute solutions)
• Robinson-Stokes (concentrated solutions)
• Bromley equation (high ionic strength)

Redox Properties

Module: microgrowagents.chemistry.redox_properties

Functions:

• calculate_redox_potential(ingredients, ph, temperature=25.0) - Calculate Eh and pE
• calculate_electron_balance(ingredients) - Calculate electron donor/acceptor balance

Constants:

• Standard redox potentials (E0' at pH 7)
• Electron equivalents for common compounds

Nutrient Ratios

Module: microgrowagents.chemistry.nutrient_ratios

Functions:

• calculate_cnp_ratios(ingredients) - Calculate C:N:P ratios and limiting nutrients
• calculate_trace_metal_ratios(ingredients) - Calculate trace metal requirements
• parse_elemental_composition(formula) - Parse chemical formulas

References:

• Redfield ratio (marine): C:N:P = 106:16:1
• Terrestrial microbes: C:N:P ≈ 60:7:1

Thermodynamic Properties

Module: microgrowagents.chemistry.thermodynamic_properties

Functions:

• calculate_gibbs_free_energy(reactants, products, ph=7.0) - Calculate ΔG
• calculate_formation_energy(compound) - Calculate ΔGf°

Data Sources:

• eQuilibrator API (biochemical thermodynamics)
• Component Contribution method
• pH and ionic strength corrections

Repository Structure

• docs/ - MkDocs documentation
  • AGENTS_SKILLS_TOOLS.md - Complete reference for all agents, skills, and tools
  • OPTIMIZATION_GUIDE.md - Complete guide to data-driven v14 design
  • OPTIMIZATION_QUICK_REFERENCE.md - One-page command reference
  • EXPERIMENTAL_INTERPRETATION_AGENT.md - Evidence-based interpretation
  • ARTIFACT_CLEANUP_POLICY.md - Retention policies and cleanup
  • AUDIT_REPORT_BBOP_SKILLS.md - Audit compliance report (1,579 lines)
  • STATUS.md - Current project state (start here)
• src/microgrowagents/ - Source code
  • agents/ - Agent implementations
  • chemistry/ - Chemistry calculation modules
  • database/ - Database utilities
  • api_clients/ - External API clients
  • skills/ - Claude Code skills framework
• tests/ - Pytest test suite (86 tests, >90% coverage)
• scripts/ - Integration and analysis scripts
  • doi_validation/ - DOI validation scripts
  • doi_corrections/ - DOI correction utilities
  • pdf_downloads/ - Automated PDF retrieval
  • enrichment/ - Data enrichment
  • schema/ - Schema management
• data/ - Database and cache files
  • raw/ - Source data with checksums
  • corrections/ - DOI correction definitions
  • results/ - Validation and processing logs
• notes/ - Research notes and documentation (27+ files)
• .claude/ - Claude Code configuration
  • provenance/ - Session manifests and action logs
  • skills/ - Claude Code skills definitions

Development

Running Tests

# Run all tests
just test

# Run specific test file
uv run pytest tests/test_chemistry/test_osmotic_properties.py -v

# Run with coverage
uv run pytest --cov=microgrowagents --cov-report=html

Type Checking

just mypy

Code Formatting

just format

Documentation

# Serve documentation locally
just _serve

# Build documentation
mkdocs build

Documentation Website

https://CultureBotAI.github.io/MicroGrowAgents

Test Coverage

• Osmotic Properties: 21/21 tests, 20 doctests
• Redox Properties: 27/27 tests
• Nutrient Ratios: 27/27 tests
• Sensitivity Analysis: 11/11 integration tests
• Total: 86 tests passing across all modules

Tools, APIs & Datasets

MicroGrowAgents integrates multiple external tools, APIs, and datasets for comprehensive microbial cultivation analysis.

External APIs

Chemical Data:

• PubChem - Chemical structure and property data, molecular formulas, identifiers
• ChEBI - Chemical Entities of Biological Interest ontology (DOI: 10.1093/nar/gkv1031)
• eQuilibrator - Biochemical thermodynamics, Gibbs free energy calculations

Biological Databases:

• KEGG - Pathway definitions, biosynthesis pathways (DOI: 10.1093/nar/gkac963)
• BRENDA - Enzyme information, EC-to-cofactor relationships (DOI: 10.1093/nar/gky1048)
• ExplorEnz - Enzyme Commission nomenclature (DOI: 10.1093/nar/gkn582)
• UniProt - Protein sequences and functional annotations
• NCBI - Genome sequences, taxonomy, literature (PubMed)

Specialized Tools (Planned):

• NIST WebBook - Inorganic thermodynamic data

Knowledge Graphs & Datasets

KG-Microbe (Primary Knowledge Graph):

• 1.5M nodes, 5.1M edges - Comprehensive microbial knowledge integration
• 864,363 validated species - Bacteria, archaea, fungi, protozoa
• Sources: GTDB (Genome Taxonomy Database), LPSN (List of Prokaryotic names), NCBI Taxonomy
• Content: Organism metadata, growth requirements, media formulations, enzyme-substrate relationships

Genome Annotations:

• 57 Bakta-annotated genomes - 667,502 features total
• Includes: Methylorubrum extorquens AM1, Methylococcus capsulatus, other model organisms
• Features: EC numbers, GO terms, gene products, cofactor requirements, transporter systems

Chemical Embeddings:

• 208,000+ chemical embeddings - Morgan fingerprints and molecular descriptors
• Use: Analogy-based reasoning, chemical similarity search, alternative ingredient discovery

MP Medium Database:

• 158 ingredients - Complete MP medium ingredient properties
• 68 columns - 47 data properties + 21 organism context fields
• 158 unique DOIs - 90.5% citation coverage (143/158 with evidence)
• 92 PDFs, 44 abstracts - Full-text evidence for ingredient recommendations

Literature Corpus:

• 245+ papers - Microbial cultivation and growth media design
• Extended information sheets - Structured metadata extraction
• Full-text search - PDF evidence extraction and excerpt retrieval

External Software Tools

Metabolic Modeling:

• GapMind - Metabolic pathway gap analysis (Morgan Price lab)
• GEMsembler - Genome-scale metabolic model reconstruction
• COBRApy - Constraint-based reconstruction and analysis (FBA)

Genome Annotation:

• Bakta - Rapid & standardized bacterial genome annotation
• NCBI BLAST - Sequence similarity search

Experimental Design:

• MaxPro OptBlock - Maximum projection optimal blocking design (custom implementation)
• Latin Hypercube Sampling - Space-filling experimental designs

Growth Prediction:

• GrowthCodon - Codon usage bias-based growth prediction
• MediaDive - Media database and search tool

Python Libraries & Dependencies

Core Scientific Computing:

• numpy - Numerical operations
• pandas - Data manipulation and analysis
• scipy - Statistical functions, optimization
• scikit-learn - Machine learning (GP regression, Random Forest, PCA)

Chemistry & Thermodynamics:

• rdkit - Chemical informatics and molecular fingerprints
• equilibrator-api - Biochemical thermodynamics

Visualization:

• matplotlib - Plotting and visualization
• seaborn - Statistical visualization
• plotly - Interactive dashboards

Database & Knowledge Graphs:

• duckdb - Embedded analytical database
• sqlalchemy - Database ORM
• linkml - Linked data modeling language

Optimization & Modeling:

• scikit-optimize - Bayesian optimization
• SALib - Sensitivity analysis (Sobol indices)
• statsmodels - Statistical modeling and ANOVA

Development:

• pytest - Testing framework
• mypy - Static type checking
• ruff - Linting and formatting
• uv - Fast Python package manager

Data Provenance

All datasets and tools are properly cited and documented:

• See data/raw/mp_medium_ingredient_properties.csv for ingredient data with DOI citations
• See docs/STATUS.md for citation coverage metrics
• See notes/DOI_CORRECTIONS_FINAL_UPDATED.md for DOI validation and corrections
• See docs/cofactor_data_sources.md for cofactor analysis sources

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Write tests for new functionality
4. Ensure all tests pass (just test)
5. Submit a pull request

License

BSD 3-Clause License. See LICENSE for details.

Copyright (c) 2026 Marcin P. Joachimiak, Lawrence Berkeley National Laboratory

Credits

This project uses the template monarch-project-copier

Citation

If you use MicroGrowAgents in your research, please cite this repository.

Contact

Principal Investigator: Dr. Marcin P. Joachimiak

• Institution: Lawrence Berkeley National Laboratory
• Project: CultureBotAI Initiative
• GitHub: CultureBotAI

For questions or issues:

• Open an issue on GitHub Issues
• See CLAUDE.md for development guidance