Computational Tumor Board (CTB)

Evolutionary Modeling Framework for Adaptive Cancer Therapy

Precision oncology treats tumors as static genomic targets. The CTB reframes cancer therapy as dynamic control of an evolving ecosystem under partial observation.

Overview

The Computational Tumor Board (CTB) is a four-layer decision-support framework that integrates:

1. Genomic targeting — Clonality-weighted molecular matching score
2. Clonal hierarchy — CCF estimation and truncal/subclonal classification
3. Evolutionary simulation — Lotka-Volterra dynamics with Stackelberg policy optimization
4. ctDNA monitoring — Closed-loop recalibration using longitudinal liquid biopsy

This repository contains the companion software for the paper:

The Oncologist as the Third Player: Evolutionary Game Theory, Precision Molecular Matching, and the Case for Computational Tumor Boards. (In preparation)

⚠️ Disclaimer

This software is a research prototype for academic use only. It is NOT a medical device and has NOT been evaluated by any regulatory authority. All outputs are model-suggested strategies intended to support, not replace, clinical judgment.

Repository Structure

ctb-repo/
├── ctb/                          # Core Python package
│   ├── __init__.py               # Package exports
│   ├── lotka_volterra.py         # ODE model with full parameter provenance
│   ├── policies.py               # Treatment policy implementations
│   ├── matching_score.py         # Molecular matching + CCF estimation
│   └── optimizer.py              # Stackelberg grid-search policy selector
│
├── notebooks/                    # Reproducible analysis notebooks
│   ├── 01_figure2_dynamics.py    # Figure 2: MTD vs AT50 vs CTB
│   ├── 02_figure3_failure_modes.py  # Figure 3: When evolutionary control fails
│   └── 03_figure5_virtual_cohort.py # Figure 5: 500-patient benchmark
│
├── data/
│   ├── parameters/
│   │   └── parameter_provenance.json  # Every parameter with published source
│   ├── outputs/                  # Generated data (reproducible from notebooks)
│   └── published_references/     # Reference PDFs / data from cited papers
│
├── tests/
│   └── test_ctb.py              # Unit + integration tests (pytest)
│
├── docs/
│   ├── METHODS.md               # Full Methods section for the paper
│   ├── MATHEMATICAL_ARCHITECTURE.md  # Equations + pipeline specification
│   └── POMDP_FRAMING.md         # Partially observed control interpretation
│
├── figures/                     # Generated figures (from notebooks)
├── requirements.txt
├── LICENSE                      # MIT
└── README.md                    # This file

Reproducibility

All data is generated, not imported

Every number in every figure comes from the Lotka-Volterra ODE model implemented in ctb/lotka_volterra.py. There are no external datasets, no pre-computed results, and no hidden data sources.

Audit trail

• Parameters: Every model parameter has a published source documented in data/parameters/parameter_provenance.json
• Random seeds: All stochastic elements (virtual patient sampling) use fixed seeds (seed=42)
• Notebooks: Each figure has a dedicated notebook that generates it from scratch
• Tests: pytest tests/test_ctb.py validates all core computations

To reproduce all results

# Install dependencies
pip install -r requirements.txt

# Run tests
pytest tests/test_ctb.py -v

# Reproduce figures
cd notebooks
python 01_figure2_dynamics.py
python 02_figure3_failure_modes.py
python 03_figure5_virtual_cohort.py

Key Results

Metric	MTD	AT50	CTB
Median TTP (days)	755	1054	>1500
Cumulative dose (% of MTD)	100%	91%	50%
Resistant fraction at endpoint	100%	100%	<1%

Virtual cohort (n=500): No single fixed policy was universally optimal. The CTB achieved the highest median utility (0.65) by adapting its strategy to each patient's ecological regime.

Parameter Sources

Parameter	Value	Source
r_S (sensitive growth)	0.0278 day⁻¹	Zhang et al., Nat Commun 2017
r_R (resistant growth)	0.02 day⁻¹	Cost of resistance assumption
α_RS (competition)	1.5	Strobl et al., Cancer Res 2024
d_S (drug kill rate)	0.018 day⁻¹	Calibrated to abiraterone response

Full provenance: see data/parameters/parameter_provenance.json

References

1. Zhang J et al. Nat Commun. 2017;8:1816.
2. Zhang J et al. eLife. 2022;11:e76284.
3. Strobl MAR et al. Cancer Res. 2024;84(11):1929.
4. Salvioli M et al. Dyn Games Appl. 2025;15:1750-1769.
5. Sicklick JK et al. JCO. 2026. (I-PREDICT)
6. Staňková K et al. JAMA Oncol. 2019;5(1):96-103.

License

MIT License. Copyright (c) 2026 Raphael Brandão.

Contact

Raphael Brandão, MD PhD(c) Medical Coordinator of Oncology, Rede São Camilo, São Paulo Founder, First Oncologia