VioFlux

Epigenetic flux simulator for the violacein biosynthesis pathway 1st place — TurbioHacks Global Bio × AI Hackathon

---

Biology

Violacein is a bisindole pigment from Chromobacterium violaceum with antibiotic and anticancer activity. Its biosynthesis proceeds through a five-enzyme linear chain:

L-Trp ──(VioA,VioB)──► IPA imine ──(VioE)──► protodeoxyviolaceinic acid
       ──(VioD)──► protoviolaceinic acid ──(VioC)──► violacein

Two competing branches set hard performance ceilings:

• CPA diversion — when VioE activity is very low, the IPA imine dimerizes spontaneously to chromopyrrolic acid (CPA), yielding zero violacein
• Chromoviridans shunt — when VioC is weak relative to the rest of the pathway, flux diverts to deoxyviolacein / chromoviridans

VioFlux simulates how epigenetic perturbations (CRISPRa, CRISPRi, methylation locks) propagate through this chain and affect yield, accounting for both the bottleneck structure and metabolic/proteome costs.

---

Model

This is a steady-state, dimensionless flux model, not an ODE. The yield score Y is a relative titer estimate under a given epigenetic configuration.

1 — Epigenetic module response

Each gene is assigned a module at a "level" (0–1, representing inducer or guide-RNA strength). The module maps level → fold-change over baseline:

Module	Formula	Boundaries
CRISPRa (activator)	1 + (A_max−1) × Hill(level, EC₅₀, h)	level=0 → 1×; level=1 → A_max×
CRISPRi (repressor)	(min_fold+leak) + (1−min_fold−leak) / (1+(level/EC₅₀)^h)	level=0 → 1×; level→∞ → min_fold+leak
Methylation (binary)	1.0 if level ≥ 0.5 else leak	ON: full expression; OFF: only constitutive leak
Neutral	1.0	no effect

Parameters (A_max, min_fold, EC₅₀, h, leak) are stored in epi-sim/data/epigenetic_modules.csv.

2 — Effective enzyme activities

a_i = k_i × F_i, where k_i is the baseline activity weight from violacein_pathway.tsv.

3 — Bottleneck flux (soft-min)

Pathway flux is limited by the slowest enzyme. A differentiable proxy for min(a):

F_core = ( mean_i( a_i^(−p) ) )^(−1/p)      p = 6  (→ min as p → ∞)
F      = min(F_core, supply_cap)

The N-normalized power mean is used so equal activities yield F_core = 1.0 at baseline (unnormalized versions carry an artifact scaling of N^(1/p)).

4 — Penalty terms

Two multiplicative penalties discount raw flux:

• Imbalance penalty P — large upstream activities relative to F_core signal metabolic buildup and potential toxicity: P = exp(−α × Σ max(0, a_i/F_core − s))

• Burden penalty Q — over-expressing all enzymes strains the E. coli proteome: Q = 1 / 1 + γ × (∑ w_i × a_i − N)²)

5 — Violacein-specific gating rules

• a[VioE] < 0.10 → hard zero (CPA diversion; no violacein produced)
• VioC weak relative to pathway median → fractional shunt toward chromoviridans (up to 35% flux loss)

Final yield

Y = F × P × (1 − shunt_leak) × Q

Gene activities are shown normalized to the pathway maximum. The Streamlit app additionally computes log-space finite-difference elasticities (analogous to MCA flux control coefficients) per gene.

---

Repository structure

VioFlux/
├── epi-sim/
│   ├── app.py                      # Streamlit UI
│   ├── sim_core.py                 # Simulation engine
│   ├── data_io.py                  # Data loaders
│   └── data/
│       ├── violacein_pathway.tsv   # Gene order, baseline k, burden weights
│       └── epigenetic_modules.csv  # Module response-curve parameters
├── web-interface/
│   └── index.html                  # Self-contained browser interface (no backend)
└── requirements.txt

---

Quick start

Streamlit app (full simulation + sensitivity analysis):

pip install -r requirements.txt
cd epi-sim
streamlit run app.py

Web interface (no install, fully offline):

open web-interface/index.html

The web interface runs the same core simulation engine in JavaScript — identical math to sim_core.py.

---

Data formats

violacein_pathway.tsv

Column	Description
gene	Gene name (VioA–VioC)
step_order	Reaction order in the chain (1–5)
baseline_k	Baseline enzyme activity (dimensionless)
burden_w	Proteome burden weight

epigenetic_modules.csv

Column	Description
module	Module key (used in UI selectors)
type	activator / repressor / binary
A_max	Max fold-activation (activator only)
min_fold	Residual expression under full repression
EC50	Half-maximal inducer concentration (0–1 normalized scale)
h	Hill coefficient (cooperativity)
leak	Constitutive transcription leak

To sub in a new pathway, replace the TSV and CSV — no code changes required.

---

Limitations

• Parameters are illustrative, not calibrated. All baseline k values are 1.0; real enzyme rates differ by orders of magnitude and are organism- and condition-dependent.
• No dynamics. This is a steady-state score. Gene expression lag, intermediate accumulation, and feedback are not modeled.
• Sensitivities are finite-difference approximations, not formal MCA flux control coefficients — the summation theorem is not enforced.
• Grid search fixes modules to CRISPRa. The Streamlit grid sweeps expression levels across {low, mid, high} for each gene but does not sweep over module types.
• VioE and VioC thresholds are heuristic. The CPA diversion cutoff and chromoviridans shunt fraction are biologically motivated but not fit to experimental data.