Navigo-release

Generative Modeling of Mouse Embryogenesis for Fate and Disease Prediction

Overview | Highlights | Installation | Tutorials | Repository Structure | Citation

Overview

Navigo is a generative modeling framework for developmental biology. It combines population-level flow matching with molecular-level RNA kinetics to learn continuous developmental vector fields from single-cell transcriptomic data.

In this repository, Navigo is packaged together with tutorial notebooks and analysis resources used for:

• developmental trajectory interpolation and denoising;
• congenital heart disease regulatory analysis;
• zero-shot prediction of genetic perturbation effects;
• fibroblast reprogramming analysis and transcription factor screening.

Navigo is designed for modeling embryogenesis across large temporal single-cell atlases and for supporting in silico perturbation and reprogramming studies.

Highlights

• Learns developmental dynamics from temporal single-cell snapshots with a continuous vector field formulation.
• Couples flow matching with transcriptional kinetics to keep the learned dynamics biologically grounded.
• Supports counterfactual perturbation analysis without perturbation-specific training.
• Enables regulatory network analysis for developmental disease settings such as congenital heart defects.
• Supports screening of reprogramming interventions, including inhibitory factors for cardiac conversion and synergistic transcription factor combinations for neuronal conversion.

Installation

Clone the repository and install the package in editable mode:

git clone https://github.com/aristoteleo/Navigo-release.git
cd Navigo-release
pip install -r requirements.txt
pip install -e .

To build the documentation locally:

pip install -r docs/requirements.txt
sphinx-build -b html docs docs/_build/html

Tutorials

The tutorial site is organized into five sections:

• Training Demo: compact end-to-end subset training and validation workflow.
• Interpolation: held-out interpolation benchmarking and denoising examples.
• GRN: congenital heart disease regulatory analysis workflow.
• Knockout: zero-shot perturbation prediction and enrichment summaries.
• Reprogramming: cardiac and neuronal reprogramming studies.

The underlying notebooks live under docs/tutorials/notebooks, and supporting resources remain under docs/tutorials/resources.

To run the tutorials locally, download the tutorial asset bundles referenced in docs/tutorials/index.md, then extract them at the repository root so data/ and checkpoints/ are available alongside docs/ and navigo/.

Repository Structure

• navigo/: installable Python package with the core modeling and analysis code.
• docs/: Sphinx/MyST documentation source for the main site and tutorial pages.
• docs/tutorials/notebooks/: end-to-end tutorial notebooks used throughout the documentation.
• data/ and checkpoints/: local tutorial assets expected by the notebooks and training demo.
• submission/: command-line entrypoints and helper scripts, including the docs sync utility.

Citation

If you use this code, please cite:

@article{fan_navigo_2026,
  title={Generative Modeling of Mouse Embryogenesis for Fate and Disease Prediction},
  author={Yimin Fan and Xinyuan Liu and Yixuan Wang and Zehua Zeng and Lei Li and Xiaojie Qiu and Yu Li},
  year={2026}
}