AcceleratingReturns: Completion Theory for Gene Panel Design

Code for: Accelerating Returns in Gene Panel Design: A Completion Theory for Regulatory Network Identification
BioSystems (Elsevier), 2026.

Repository Structure

Parsing

• parsing/parse_bbm.py — BBM database cloning + .bnet parser (parse_bnet(), parse_all())

Core Algorithms

• core/hypergraph.py — Identifiability hypergraph (Definition 1), greedy/exact hitting sets, coverage functions
• core/strategies.py — Panel selection: corollary_order() (Algorithm 1, α=0.25), two_phase_greedy_order(), degree_order(), random_order()
• core/theorem.py — Theorem 3: predicted completion E[Δ(k)], cumulative curves, acceleration ratios
• core/utils.py — Cumulative completions, observability decomposition (Proposition 1), degree-preserved null model

Revision Experiments

• revision/experiments.py — Six experiments addressing reviewer concerns:
  • experiment_1() — Protocol 2: realistic observation budgets (coupon-collector identifiability)
  • experiment_2() — Dropout robustness (p_drop ∈ {0.05, 0.10, 0.20, 0.30, 0.50})
  • experiment_3() — Computational scalability (synthetic networks N=50–2000, power-law fit)
  • experiment_4() — Full 285-network comparison (expands Table 1 from n=50)
  • experiment_5() — Error-tolerant (ε-relaxed) completion
  • experiment_6() — Theorem prediction vs algorithm performance

Reproducibility

• reproduce_all.py — Regenerate all results from parsed networks
• generate_figures.py — Regenerate all figures (7 paper + 7 revision, 600 DPI)
• requirements.txt — Pinned dependency versions

Checkpoints, Results, and Data

Available at: https://huggingface.co/Laddaphone/accelerating-returns-panel-design

• data/parsed_networks.pkl — 285 parsed Boolean regulatory networks (N=5–1076)
• 6 revision experiment checkpoints (results/revision/)
• 15+ original experiment checkpoints (results/original/)
• 7 paper figures as PDF (figures/paper/)
• 7 revision figures as PDF+PNG (figures/revision/)

Quick Start

pip install -r requirements.txt
python reproduce_all.py --data_dir ./data --output_dir ./results
python generate_figures.py

Key Results

Result	Value
Theorem validation (ρ, 60 networks)	0.9998 ± 0.0011
Corollary vs greedy at k/N=0.5 (n=285)	+12.3 pp (p < 10⁻⁴³)
Corollary vs random at k/N=0.5 (n=285)	+32.2 pp (d = 3.67)
Protocol 2 threshold shift at m=100	+5.6 pp
Dropout robustness at p=0.20	Advantage persists (p = 10⁻¹¹)
Scaling exponent	O(N³·¹⁵)

Citation

Douangnouanexay, L. (2026). Accelerating Returns in Gene Panel Design: A Completion Theory for Regulatory Network Identification. BioSystems.

License

MIT