myconet 1.1.0

Python framework for mycorrhizal network biophysics and mathematical biology transport analysis, with Freiman–Villani thermodynamic diagnostics

myconet

Python simulation framework for mycorrhizal network biophysics and general mathematical biology transport analysis, with Freiman–Villani thermodynamic diagnostics.

---

Overview

myconet implements a suite of mathematical tools originally developed for the analysis of mycorrhizal (fungal root) networks, grounded in Villani's hypocoercivity theory, the Freiman–Ruzsa structural theorem, and optimal transport (Wasserstein-2 / Sinkhorn). The library has since been applied as a unified transport diagnostic framework across a family of multiscale mathematical biology models for neurological and respiratory diseases.

Core mathematical backbone:

• Wasserstein-2 distance (Sinkhorn algorithm via POT) — measures divergence of a cell/network population distribution from a healthy reference
• Relative entropy H(ρ|ρ∞) — KL divergence from the steady state
• Fisher information I(ρ|ρ∞) — metabolic dissipation proxy
• Freiman excess σ_r — structural order relative to hexagonal reference
• Villani dissipation lower bound Ψ_lb — hypocoercivity-based convergence rate
• Talagrand T₂ lower bound — saturation ratio ρ_T = λ_hc·W₂²/(2H)
• Fokker–Planck PDE solver — mesoscale population transport

---

Installation

pip install myconet

Requirements: numpy, scipy, matplotlib, POT (Python Optimal Transport), scikit-learn

---

Core API

from myconet.transport import wasserstein2, relative_entropy, fisher_information
from myconet.freiman  import dissipation_lower_bound, w2_lower_bound, freiman_excess
from myconet.network  import HyphalNetwork, hexagonal_lattice
from myconet.simulation import MycoNetSimulation, SimulationParams, run_ensemble

Transport diagnostics

import numpy as np
from myconet.transport import wasserstein2, relative_entropy, fisher_information

# nodes: (N, 2) array — current population distribution in R²
# ref  : (M, 2) array — healthy reference distribution
# eps  : float        — characteristic scale

nodes = np.random.rand(100, 2)          # current state (e.g. cell positions)
ref   = np.random.rand(100, 2) * 0.3   # healthy reference

W2     = wasserstein2(nodes, ref, eps=0.30)       # Sinkhorn W₂
H      = relative_entropy(nodes, ref, eps=0.30)   # KL divergence
I_fish = fisher_information(nodes, ref, eps=0.30) # Fisher information

Freiman–Villani diagnostics

from myconet.freiman import freiman_excess, w2_lower_bound, dissipation_lower_bound

sigma_r = freiman_excess(nodes, eps=0.30)              # structural excess
W2_lb   = w2_lower_bound(sigma_r, eps=0.30)            # Talagrand T₂ lower bound
Psi_lb  = dissipation_lower_bound(sigma_r, eps=0.30, D=0.05)  # Villani dissipation

---

Applications

Original application — Mycorrhizal network biophysics

myconet was developed for the companion paper:

Mercier des Rochettes, B. (2026). Geometric Efficiency Bounds for Mycorrhizal Networks: A Freiman–Villani Framework. Journal of Mathematical Biology. myconet v1.0.0.

The core result: mycorrhizal hyphal networks self-organise toward the hexagonal lattice (K_HEX ≈ 0.6412) as the optimal transport configuration. The Freiman excess σ_r − K_HEX quantifies deviation from this optimum, and the HWI inequality W₂² ≤ (2/λ_hc)H bounds convergence.

---

Biomedical applications (v1.1.0)

Starting from v1.1.0, myconet includes four complete multiscale disease models as worked examples. Each applies the Villani/HWI/Talagrand framework to a different biological system via a 2D embedding of the cellular state space.

Embedding principle: the 8-dimensional cellular state is projected onto two biologically meaningful planes — an inflammatory/pathological plane and a functional/repair plane — represented as Gaussian point clouds. The myconet transport functions then compute the divergence of this distribution from a healthy reference.

Example	Disease	Inflammatory plane	Functional plane	Clinical readout
ms_multiscale	Multiple Sclerosis	(Teff, M1)	(OL, My)	EDSS, MRI lesion load
ad_multiscale	Alzheimer's Disease	(M1_micro, A1_astro)	(N_neu, N_syn)	MMSE, Aβ/tau PET
pd_multiscale	Parkinson's Disease	(DAQ, M1_micro)	(DA_neu, DA_level)	UPDRS-III, DAT-SPECT
cf_multiscale	Cystic Fibrosis	(Pa_biofilm, Neutrophil)	(CFTR_rescue, ASL_norm)	FEV1%, Pa index

Each example is a self-contained four-scale model:

• L0 Molecular: Nakajima–Zwanzig non-Markovian memory kernels (Prony form)
• L1 Aggregation: Smoluchowski coagulation-fragmentation (disease-specific)
• L2 Cellular: ODE system + myconet Villani/HWI/Talagrand diagnostics
• L3 Clinical: Cusp catastrophe + validated clinical score

See examples/ for full code and documentation.

---

Mathematical background

HWI inequality (Villani 2009)

For a probability measure ρ on ℝ² with λ_hc-log-Sobolev inequality:

$$W_2^2(\rho, \rho_\infty) \leq \frac{2}{\lambda_{hc}} H(\rho ,|, \rho_\infty)$$

where W₂ is the Wasserstein-2 distance, H is the relative entropy (KL divergence), and λ_hc is the hypocoercivity rate (spectral gap of the linearised Fokker–Planck operator).

Talagrand T₂ saturation ratio

$$\rho_T = \frac{\lambda_{hc} \cdot W_2^2}{2 H(\rho|\rho_\infty)} \in (0, 1]$$

ρ_T → 1 indicates the HWI bound is tight (population at thermodynamic equilibrium with the healthy reference). ρ_T < 1 indicates remaining distance to be covered — usable as a treatment response biomarker.

Freiman excess

$$\sigma_r = \frac{|A + A|}{|A|} - K_{HEX}$$

where A is the point set of network nodes, |A+A| is the sumset cardinality, and K_HEX ≈ 0.6412 is the hexagonal packing constant. σ_r < 0 indicates over-ordered (pathological) polarisation; σ_r → 0 indicates convergence to the hexagonal optimum.

---

Citation

If you use myconet in your research, please cite:

@software{myconet2026,
  author    = {Mercier des Rochettes, Bertrand},
  title     = {myconet: Python framework for mycorrhizal network biophysics
               and mathematical biology transport analysis},
  year      = {2026},
  version   = {1.1.0},
  publisher = {PyPI},
  url       = {https://pypi.org/project/myconet/},
  note      = {Quantum Proteins AI, Cergy-Pontoise, France}
}

For the biomedical applications, please also cite the corresponding JMB papers (references to be updated on acceptance):

• MS model: Mercier des Rochettes (2026, JMB, submitted)
• AD model: Mercier des Rochettes (2026, JMB, submitted)
• PD model: Mercier des Rochettes (2026, JMB, submitted)
• CF model: Mercier des Rochettes (2026, JMB, submitted)

---

Changelog

See CHANGELOG.md.

---

License

MIT — see LICENSE.

Author: Bertrand Mercier des Rochettes
Institution: Quantum Proteins AI, Cergy-Pontoise, France
Contact: [e-mail]
Website: https://quantum-proteins.ai