Python toolkit and Streamlit apps for fermentation analysis — growth kinetics, elemental mass balance, scale-up modeling, and a digital twin — validated against literature datasets spanning E. coli, B. subtilis, C. glutamicum, K. phaffii/P. pastoris, P. putida, L. lactis, S. cerevisiae, Y. lipolytica and CHO.
A multi-page Streamlit app built around ferment.py, an analysis engine with automatic phase detection (adaptive thresholds), corrected μ_max calculation across growth regimes, and fed-batch / high-cell-density handling.
| Page | What it does |
|---|---|
| Data Import | Universal parsing (CSV/TSV/JSON, Eppendorf/Infors/Sartorius exports) with full sensor column mapping; organized run storage and metadata via data_manager.py |
| Analysis | Combined growth, substrate, product and sensor profiles on one timeline, with lag/exponential/stationary phase overlay |
| Growth Kinetics | μ, doubling time, lag, productivity, R², Luedeking–Piret growth/non-growth-associated product model, literature comparison per organism/medium |
| Mass Balance | Multi-substrate, multi-product C/H/O/N elemental balance, degree-of-reduction checks, and thermodynamic impossible-yield detection |
| Scale-up | Full engineering model: kLa (van't Riet), mixing time (Nienow), P/V, tip speed, Reynolds number — with a literature validation dataset |
| Digital Twin | Self-learning Monod-ODE fit and prediction for batch, fed-batch and continuous modes |
| Product & Metabolite Profiles | Time-course multi-metabolite plots, assay integration, anomaly and contamination detection |
| Golden Batch | KPI comparison against a reference run with automated explanations and a radar chart |
| Strain Library | Glycerol bank (cryostock), passage history, shake-flask pre-culture stages |
| Run Registry | Project, client, objective, outcome, KPIs, analytics and timeline per run |
| AI Assistant | Context-aware natural-language Q&A grounded in the currently loaded run data (ANTHROPIC_API_KEY in env; never committed) |
| Dashboard | Central hub for stored runs and fermentation data management |
Validation: test_with_literature.py and auto_validate.py benchmark the calculations against published values for each organism; 30+ literature-derived and synthetic multiprobe datasets ship in the repo (see MULTIPROBE_DATASET_README.md).
| Path | What it is |
|---|---|
analytical_databank/ |
Structured reference of biotech/biopharma analytical instruments — principle, real industry models, methods by product type (mAbs, vaccines, cell & gene therapy), report contents and acceptance criteria, ICH/USP/EP references |
synapse.py + synapse_pages/ + pulse.py |
SYNAPSE: the unified platform that merges the analyzer and the databank — run manager, live-run view with cross-referenced instrument methods, batch history with trend charts, instrument reference browser, and Pulse, a telemetry writer that streams live readings to Supabase every 60 s during an active batch (supabase_schema.sql; credentials in .env, never committed) |
gem-map-builder/ |
Maps analyzer readings onto genome-scale metabolic model pathways for a live metabolic overlay — 10 production organisms across bacterial, yeast and mammalian platforms, with reaction validation and health scoring |
sample_data/ |
Small curated E. coli batch/fed-batch datasets for a quick first run |
cd bpa_app
pip install -r requirements.txt
streamlit run app.pyEverything on this path runs fully offline — load any dataset from sample_data/ or the organism CSVs in bpa_app/. For the full SYNAPSE platform: pip install supabase python-dotenv, add Supabase credentials to .env, and streamlit run synapse.py.
 |
 |
 |
 |
 |
 |
Growth Kinetics on the bundled E. coli MG1655 glucose dataset: μ = 0.448 ± 0.006 h⁻¹ (95% CI), R² = 0.9993, doubling time benchmarked against literature values.
I spent nine years developing and scaling fermentation processes, including a 1,000 kg production campaign at a contract manufacturer. This is the analysis layer I always wanted at the bench: the calculations a process scientist actually runs — kinetics, mass balance, scale-down/scale-up criteria — with the sanity checks (elemental closure, thermodynamic yield limits, phase-aware μ) that catch bad data before it drives a bad decision.
— Anupama Kozhiyalam · github.com/anupama29k