dendra 0.27.2

Dendra is a differentiable biophysical neuron and network simulator built on PyTorch. It supports CPU and GPU execution, extracellular fields, event-based networks, and macroscopic descriptors for gradient-based model training.

Documentation

The complete documentation is published at wmglab-duke.github.io/dendra. Start with the installation guide, then see the tutorials and the unit conventions.

Installation

Dendra requires Python 3.11 or newer and PyTorch 2.12 or newer. On Windows, install inside WSL2 because the required NEURON dependency does not publish native Windows wheels on PyPI.

Install Dendra with the recommended native CPU solvers:

python -m pip install --upgrade pip
python -m pip install --only-binary=dendra-solvers "dendra[solvers]"

The solvers extra installs dendra-solvers, which provides native CPU solvers for block and tree models. If a compatible solver wheel is unavailable, install the base package:

python -m pip install dendra

The base package retains Dendra's PyTorch CPU solver for unbranched cables and all GPU solvers. To select a particular CPU, CUDA, or ROCm build, install PyTorch first using its installation selector, then install Dendra.

For interactive Matplotlib figures in Jupyter, install the Jupyter extra:

python -m pip install "dendra[jupyter]"

Verify the installed dependencies and inspect the deployment environment:

python -m pip check
python -m dendra doctor

The installation guide explains solver-wheel support, source installations, upgrades, and Jupyter setup. For CUDA deployment checks, see the GPU deployment guide.

Development

The contribution guide describes development setup and the GitHub pull-request workflow. The testing and code coverage guide covers the CPU, NEURON, CUDA, sanitizer, and coverage lanes.

Citation

A citation for Dendra and its accompanying manuscript will be added when it is available. Please also consider citing the work where geometric and topological surrogates and gradient-based design of selective neurostimulation are discussed in detail:

Minhaj A. Hussain, Warren M. Grill, Nicole A. Pelot. "Highly efficient modeling and optimization of neural fiber responses to electrical stimulation." Nature Communications. 2024. doi:10.1038/s41467-024-51709-8

@article{hussain_highly_2024,
    title = {Highly efficient modeling and optimization of neural fiber responses to electrical stimulation},
    doi = {10.1038/s41467-024-51709-8},
    journal = {Nature Communications},
    author = {Hussain, Minhaj A. and Grill, Warren M. and Pelot, Nicole A.},
    year = {2024}
}

License

Dendra is distributed under Duke University's custom license for non-commercial research and academic testing. Commercial use, including industrially sponsored research, requires a separate agreement with Duke's Office for Translation and Commercialization. The license governs all use; downloading, cloning, or forking the software constitutes acceptance of those terms.