MFA Pipeline Orchestrator

Overview

Coordinates all mfa-agent sub-agents in sequence, tracking progress via progress/ markdown files so any failed step can be resumed independently.

Full pipeline:

Model source (BIGG ID / custom reactions)
  → [model-builder]     models/<Model>.json  +  validation report
  → [fba-runner]        simulations/fba_fluxes.csv  +  scan_summary.json
  → [flux-analyzer]     analysis/essentiality.csv  +  phase_plane.png
  → [metabolic-pheno-analyzer]  output/figures/*.pdf  +  yield table

Workflow

Step 0: Parse User Request

Extract and record in progress/step0_inputs.md:

• Model source (BIGG ID or custom)
• Organism and condition (aerobic/anaerobic, carbon source, concentration)
• Objective reaction (biomass or product)
• Gene knockouts to apply
• Analysis goals (essentiality, phase plane, yield optimisation, WT vs. mutant comparison)
• Target product (if yield analysis requested)

Step 1: Invoke model-builder

Provide: model source, medium constraints, objective, knockouts. Wait for progress/step1_metabolic_model.md. Read: model file path, WT growth rate, model statistics.

Step 2: Invoke fba-runner

Provide: model path, simulation types requested (FBA, pFBA, FVA, knockout screen), carbon source sweep if requested. Wait for progress/step2_fba_simulation.md. Read: flux CSV paths, essential gene count, secretion fluxes.

Step 3: Invoke flux-analyzer

Provide: model path, FBA results, analysis goals (essentiality, phase plane, sampling), nutrient pair for phase plane. Wait for progress/step3_flux_analysis.md. Read: essential genes, phase plane optimum, engineering targets.

Step 4: Invoke metabolic-pheno-analyzer

Provide: model path, all previous results, target product, publication requirements. Wait for progress/step4_metabolic_phenotype.md. Read: max theoretical yield, figure paths.

Progress File Specification

progress/step1_metabolic_model.md

# Step 1: Metabolic Model
## Status: PASS / FAIL
## Model: <BIGG_ID>.json
## Reactions: N  Metabolites: M  Genes: G
## WT growth rate: X h⁻¹
## Validation: mass balance errors=0, dead-ends=N

progress/step2_fba_simulation.md

# Step 2: FBA Simulation
## Status: PASS / FAIL
## Runs: FBA, pFBA, FVA, knockout screen
## WT growth rate: X h⁻¹ (pFBA: Y h⁻¹)
## Essential genes: N
## Key secretion products: [ethanol: X mmol/gDW/h, ...]
## Files: simulations/fba_fluxes.csv, simulations/gene_essentiality.csv

progress/step3_flux_analysis.md

# Step 3: Flux Analysis
## Status: PASS / FAIL
## Essential gene count: N
## Phase plane optimum: glucose=X, O2=Y → growth=Z h⁻¹
## Top engineering targets: [gene1, gene2, gene3]
## Files: analysis/phase_plane.png, analysis/essentiality.csv

Key Conventions

• Never re-run completed steps — check progress file status before invoking sub-agents
• Maximum total sub-agent retries: 10 across all steps
• All file paths relative to working directory
• The orchestrator does not run FBA itself — all computation delegated to sub-agents