Omics + organism → ranked interventions → runnable wet-lab protocol. One guided flow on the validated tools.
…or upload your own model:
…or upload your RNA-seq CSV (genes × samples):
Ranked edits come from validated OptKnock+FSEOF (recall@10 ≈0.46); the omics branch adds DEGs and reporter metabolites as insight. The protocol is a reviewed Opentrons template, not a validated run.
FluxPilot™
Your junior scientist for
Metabolic Engineering
It proposes, you decide.
Quick Actions
Upload Data
Import metabolomics, transcriptomics, and flux data
Train Model
Configure and start Enhanced MIE training
Design Experiments
AI-powered experimental design optimization
Published-Result Benchmarks
v1 — scaffold
Each case compares FluxPilot's MES-ranked interventions against a published wet-lab overproducer result.
Use these to calibrate expectations, to validate the pipeline against your own organism's known biology,
or to cite FluxPilot in publications.
Loading…
Select a benchmark on the left to view the case spec, expected canonicals,
and citation.
Spatial Benchmarks
spatially-resolved datasets
Strain-engineering benchmarks are bulk. Spatial biology is a separate vertical — tumor microenvironments, tissue zonation, bacterial-community imaging. Click to open the matching Spatial tab.
Loading…
Data Management
Upload your omics data — or run a published-result benchmark
Want to skip upload? Use a published-result benchmark — 6 organisms (E. coli, yeast, B. subtilis, C. glutamicum, Y. lipolytica, C. necator) with bundled models + real omics where available.
Demo dataset loaded — go to the Training tab and click Start Analysis
Taxon abundance profile → community FBA. Map taxa via a model column, a taxon→model file, or an AGORA database (below).
AGORA model database (optional)
🗄
Drop files or browse
AGORA/CARVEME DB (.qza/.zip) → MICOM builds each metagenomics taxon by matching this rank:
Enzyme kcat table (optional)
⚡
Drop files or browse
Per-reaction (reaction_id,kcat) or EC (ec_number,kcat) table → GECKO enzyme-capacity constraints. From a sequence-ML predictor (CatPred/TurNuP) or a BRENDA/SABIO-RK export.
GEM Model (optional) sample
SBML .xmlBiGG built-in
⌕
—✕
🧬
Or drop SBML file / browse
Uses built-in model if omitted
Raw Sequencing (FASTQ)
FASTQ.fastq.gz.fq.gz
🧪
Drop FASTQ or browse
Paired-end R1 + R2 files
Measured Fluxes (optional)sample
CSV13C MFA
📈
Drop 13C fluxomics CSV or browse
Best accuracy — replaces pFBA with measured fluxes
Measured Outcomes (optional)sample
CSVTSVXLSX
📊
Drop measurements CSV or browse
Measured growth/titer/yield per condition - the LABEL for honest ML (join key: condition_id)
Media Conditions (optional)sample
CSVBoundsMetadata
🧪
Drop conditions or metadata CSV or browse
Exchange bounds per condition, or PRECISE-1K-style metadata (carbon source + growth rate)
Low-confidence models auto-fall-back to analytical-only mode.
Use FCC Baseline Mode and Min Growth Constraint
to control how the baseline FBA state is defined.
Simulation Context
Sets FBA exchange bounds
Warm-starts ML training
0%
Complete
Initialising...
Live Log
↓ Save Log↑ Top
Waiting for analysis to start...
Training Metrics
↓ Save PNG
Analysis Results
Live results from the FluxPilot™ Decision Engine — run an analysis in the Training tab to populate
Active-flux backbone of the integrated model. Communities by colour; particle speed and edge weight scale with carbon flux. Hover a metabolite for its top reaction.
Pathway Flux Map
Top flux-carrying reactions. Intervention targets highlighted.
Interactive Metabolic Map
Escher.js -- pan, zoom, click reactions. Flux data overlaid from FBA.
Optimize your experiments with intelligent design recommendations
Design Space Exploration
AI Media Optimizer
Upload a metabolic model and let Optuna search the exchange medium to maximize growth, product flux, or both.
Optionally layer in transcriptomics for a side-by-side comparison.
Current medium (exchange reactions)
Uptake values shown in mmol/gDW/h (positive = into cell). The optimizer searches these bounds server-side.
Reaction
Metabolite
Max uptake (mmol/gDW/h)
Spatial Transcriptomics
Checking mode...
Integrate single-cell RNA-seq and spatial transcriptomics with genome-scale metabolic models
scRNA-seq Data sample
.h5ad (AnnData) or CSV — genes x cells matrix
Ready
Spatial Data sample
.h5ad with spatial coords or CSV with x,y columns
Optional — adds spatial heatmap
GEM Model sample
SBML metabolic model (.xml) — Built-in if omitted
Optional
Configuration
ProAdvanced Spatial Parameters
Clustering · QC · Context Model
▼
Tissue Metabolite Map
(region colouring by dominant metabolic state)
OmniFlux Multi-Omics Integration
Ready
Integrates all data types (bulk transcriptomics, proteomics, metabolomics, scRNA-seq, spatial transcriptomics, spatial metabolomics)
with a genome-scale model to produce spatially-resolved intervention recommendations.
Data is inherited from other tabs automatically.
Demo mode active — all 7 data sources loaded (E. coli K-12, iML1515)
Required
●
GEM Model
Select on Data tab
●
Bulk Transcriptomics
Upload on Data tab
●
Bulk Proteomics
Upload on Data tab
●
Bulk Metabolomics
Upload on Data tab
●
scRNA-seq
Upload on SpatT tab
●
Spatial TX
Upload on SpatT tab
●
Spatial Metabolomics
Upload on SpatM tab
Advanced Parameters
25 configurable options
Initialising...0%
Per-Zone Target Flux
Integration Summary
Volcano Plot — Differential Gene Expression
Global Interventions (ranked across all zones)
#
Action
Reaction
Gene Targets
Metabolites
Score
Fold Change
Confidence
Zones
Cross-Zone Metabolic Transfers
Warnings & Skipped Analyses
Workflow Builder
BETA
Drag modules from the left onto the canvas and connect their ports.
A connection turns green only when
the output type matches the input type. Set parameters on each node, then Run.
⚙ Run settingsshared across all nodes — a node's own value overrides
Recommended interventionsValidated
Fast, omics-free design — needs only a model + target. No RL training.
PHASE 1
Ingestion
Waiting
PHASE 2
Refinery
Waiting
PHASE 3
PPO Training
Waiting
PHASE 4
Report
Waiting
Initialising...0%
Waiting to start...
Flux Changes
(wild-type vs agent-optimised · sorted by |Δflux|)
Top Flux Shifts
Refinery Log
AI Assistant
BETA
Describe a metabolic-engineering goal in plain language. The assistant plans it,
runs the validated design tools (FBA baseline +
OptKnock/FSEOF), and explains the result — it never invents genes or numbers.
Try: “increase succinate in E. coli on glucose, anaerobic” or
“boost lysine yield in E. coli”.
Experiments
BETA
Your logged designs and measured outcomes — the Design-Build-Test-Learn dataset.
Log a design from the Assistant, then submit the measured lab result here to
close the loop. As results accumulate, the calibration below shows how well
FluxPilot's predictions track reality.
Validation & Limitations
Empirical recovery rates from FluxPilot's published-canonical benchmark corpus,
plus the known-failure modes documented during the T-C3.13 audit (session 50.25).
This is the page to read before relying on a prediction for wet-lab planning.
Loading validation snapshot…
EXGEM
COMING SOON
EXGEM creates, curates, and integrates multi-omics data into Metabolism + Expression (ME) models.
Unlike standard flux models, ME-models account for protein synthesis costs, ribosome allocation,
mRNA production, and transcription/translation burden — predicting how much it actually costs
the cell to express pathway enzymes.
STEP 1
ME-Model Build
Build full Metabolism + Expression model from COBRA M-model. Adds ~4,000 translation, transcription, and tRNA reactions.
STEP 2
Transcriptomics
Cap mRNA synthesis fluxes using measured gene expression. Low-expression genes get tighter bounds.
STEP 3
Proteomics
Cap protein synthesis fluxes using measured protein abundances. Directly constrains proteome capacity.
STEP 4
Metabolomics
Set metabolite concentration bounds for enzyme kinetics. Measured concentrations replace defaults.
STEP 5
EXGEM Solve
Binary search for maximum growth rate. ME growth rate is always lower than FBA due to expression costs.
STEP 6
Spatial Resolution
Solve per tissue cluster for spatially resolved protein costs and ribosome allocation maps.
What EXGEM Produces
✓ ME growth rate (expression-aware)
✓ Protein allocation budget
✓ Top expensive proteins
✓ mRNA synthesis costs per gene
✓ Ribosome fraction of proteome
✓ Expression burden (% growth loss)
✓ Target pathway protein cost
✓ Per-cluster spatial ME metrics
EXGEM Preview
Preview
EXGEM is in the final stages of development
The full ME-model integration with omics data will be available in an upcoming release. Run an analysis to see preview metrics from published reference data.
Microbiome — community metabolic modeling
Decision-support — relative & mechanistic.
This predicts the cross-feeding map (who-feeds-whom), relative per-taxon growth & SCFA capacity, and directional what-if scenarios.
It is not an absolute fecal-SCFA-concentration predictor — a genus-level validation gave ~0 correlation to absolute concentrations, so trust the
direction and mechanism, not the absolute numbers. Runs MICOM community FBA in an isolated solver environment.
0.5
Set the fraction (and optionally a community), then Run. Results: community + per-taxon growth, SCFA exchange, and the cross-feeding map.
Community (multi-microbe) from SBML models
Upload two or more SBML / JSON metabolic models. Each file is registered as a community member.
Calls /api/community for cross-feeding analysis.
Upload models and click Analyze community.
Run Comparison
Compare two completed OmniFlux or StrainPilot runs side by side.
Run A
Run B
Intervention Score Comparison
Intervention Differences
Reaction
Action A
Score A
Action B
Score B
Delta
Single-cell RNA-seq
Upload a single-cell matrix (.h5ad or genes×cells CSV). FluxPilot
clusters the cells, then builds a context-specific metabolic model for each
cell-type cluster and ranks interventions per cluster. Needs full-transcriptome
coverage — a targeted panel of a few hundred genes is too sparse to rank interventions.
1. Single-cell matrix
2. Metabolic model (SBML)
Optional — a built-in E. coli model is used if none is given.
ATAC-seq
ATAC-seq measures which DNA regions are open, not which genes are expressed.
FluxPilot collapses accessibility to a per-gene activity score and then runs the same
per-cell-type metabolic pipeline as scRNA (the approach validated against measured flux
in Cheng et al. 2022). Provide either a gene-activity matrix (genes×cells) or a
peaks×cells matrix with a peak→gene map. Raw coordinate→gene linking
belongs upstream in ArchR / Signac.
1. ATAC matrix
2. Peak→gene map (optional)
2 columns: peak_id, gene_id. Omit if the matrix is already gene-level.
3. Metabolic model
Optional — built-in E. coli model used if none.
Multiome (scRNA + scATAC)
Paired single-cell RNA and ATAC from the same cells. FluxPilot combines them
so a gene constrains its reactions only where it is both accessible
(ATAC) and expressed (RNA), then builds a metabolic model per cell-type
cluster. The two matrices must share cell barcodes and gene ids.
1. scRNA matrix
2. scATAC matrix
Gene-activity, or peaks + a peak→gene map below.
3. Model + map (optional)
model + optional peak→gene map
Curation
Everything below is something FluxPilot proposed, not something it knows.
Rejecting is as useful as accepting: a wrong answer you flag is worth more than a
right answer you wave through. Mapping and model fixes are saved against the model,
so a later run reuses them instead of asking again.
Literature grounding is off in this build — it needs an LLM key. Set one (or run Ollama) to turn it on.
Uses the model uploaded on the Data tab. Check model is instant and covers
mass balance, charge balance and GPR integrity. Deep check runs the full
MEMOTE suite, the runnable form of the Thiele & Palsson reconstruction protocol, and
takes roughly 2 to 3 minutes on a genome-scale model.
Inspect any reaction or metabolite — the checks only flag defects they can detect,
so this is how you catch a wrong direction or GPR that still balances:
Loading queue…
Run History
All your past runs in one place. Click any row to expand inline results.
ID
Status
Organism
Model
Gaps
Precision
Confidence
Duration
Date
Actions
No runs found. Start an analysis to see results here.
Copy this token (in production, this would be emailed to you):
Set New Password
Enter your reset token and new password
LLM Backend Settings
Choose your preferred AI backend and enter API keys. Keys are encrypted and stored securely.
GeminiNot set
AnthropicNot set
GroqNot set
NCBI (PubMed)Not set
Ollama runs locally — no API key needed. Ensure ollama serve is running on your machine.
Send Feedback
Rate modules you used (1-5 stars). Leave blank to skip.
General
Modules
FLUXPILOT · INTERACTIVE DEMO
Explore the FluxPilot workspace
Register to open a live, read-only tour of the interface.
A demonstration of the interface. No account is created, and nothing runs. You can browse the real published-result benchmark library, and every other action is disabled.