scMagnify's Reproducibility Repository

This repository contains code, data layout, and notebooks to reproduce benchmarks and figures for the scMagnify.

This repository accompanies the preprint Decomposing multi-scale dynamic regulatory from single-cell multiomics with scMagnify (Chen*, ..., et al., bioRxiv, 2025).

• The repository is on GitHub here
• This repository contains all code used to benchmark scMagnify and generate the figures in the manuscript.
• Jump to the Code to produce figures section for links to the specific notebooks for each figure panel.

Installation

• Python >= 3.10
• Install dependencies from pyproject.toml, e.g. with pip, If you don't have Python installed, we recommend installing uv.

uv pip install git+https://github.com/LiHongCSBLab/scMagnify.git@main

Quick start

• Re-run simulations/benchmarks:
  • benchmark/simulation/* for scMultiSim evaluations
  • benchmark/baseline/* for baseline method runs
  • benchmark/scmagnify_multirun/* for multi-run configs/outputs
• Re-generate figures: see figures/Fig2–Fig6 assets and companion notebooks.

Programmatic API (src/grn_tools):

• from grn_tools import GRNEvaluator
• evaluator = GRNEvaluator(); evaluator.load_grns(...); evaluator.load_groundtruths(...)
• evaluator.calculate_accuracy(); evaluator.plot_performance_curves('pr'|'roc', group_by='Dataset'|'Lineage')

Repository structure

• benchmark/
  • baseline/: scripts to run baseline GRN methods for comparison
  • scmagnify_multirun/: hydra-based multi-run configs/outputs for scMagnify
  • simulation/: notebooks and scripts for scMultiSim-based simulations and evaluation
  • figures/, results/: aggregated outputs from benchmarks
• data/: input data stubs and dataset folders (large files hosted externally)
• figures/: figure-specific assets (Fig2–Fig6)
• notebooks/: jupyter-book scaffolding and templates
• scripts/: small helper scripts/templates
• src/grn_tools/: evaluation utilities
  • GRNEvaluator.py: end-to-end GRN evaluation workflow
  • _acc_metrics.py: AUPR/AUROC/F-score/EPR and batch evaluators
  • _stab_metrics.py: Jaccard/Cosine stability metrics and plotting
  • _plotting.py: plotting helpers (bar/violin/line/heatmaps)
  • _utils.py: small helpers (matrix-to-edge, image merge, etc.)
  • _constants.py: dataset paths, palettes, and ground-truth maps

Codebase

This repository is meant to enhance the Materials & Methods section by providing code for all analyses in the manuscript, in order to improve reproducibility for the main results.

• scMagnify-benchmark --> All scripts and notebooks for benchmarking and simulation.
  • baseline --> Scripts for running baseline GRN inference methods (SCENIC, CellOracle, Velorama, etc.) for comparison.
  • magnify_multirun --> hydra-based scripts for executing scMagnify inference across different datasets and parameters (conf, multirun, outputs).
  • simulation --> Jupyter notebooks for generating simulated data and evaluating the performance of scMagnify against baseline methods.
• scMagnify-figures --> Jupyter notebooks organized by figure number, used to generate all main and supplementary figures in the manuscript.
  • Fig2 --> Notebooks for data preprocessing and TF binding/driver analysis.
  • Fig3 --> Notebooks for analyzing regulatory dynamics and time lags.
  • Fig4 --> Notebooks for RegFactor decomposition (NaiveB lineage).
  • Fig5 --> Notebooks for differential RegFactor analysis (Alpha vs. Beta lineage).
  • Fig6 --> Notebooks for intracellular communication (L-R to TF) analysis.

Code to produce the figures

This table contains pointers to the key notebooks associated with each figure.

Figure	Analysis	Path
Fig 1 / S1	Simulation & Benchmark Evaluation	scMagnify-benchmark/simulation/06_scMultiSim_evaluate.ipynb
Fig 2a	Data Object Preparation	scMagnify-figures/Fig2/00_data_object.ipynb
Fig 2b / S2	TF Binding Accuracy (Cistrome)	scMagnify-figures/Fig2/01_tf_binding_accuracy-Cistrome.ipynb
Fig 2c / S3	Driver TF Recovery	scMagnify-figures/Fig2/02_driver_tf_recovery.ipynb
Fig 3a	GData Object (Inference)	scMagnify-figures/Fig3/00_gdata_object.ipynb
Fig 3b	Regulatory Specificity	scMagnify-figures/Fig3/01_reg_spec.ipynb
Fig 3c	Regulatory Dynamics	scMagnify-figures/Fig3/02_reg_dyn.ipynb
Fig 4	RegFactor Analysis (NaiveB)	scMagnify-figures/Fig4/01_regfactor_analysis_NaiveB.ipynb
Fig 5a	Differential RegFactor Analysis	scMagnify-figures/Fig5/01_reg_diff.ipynb
Fig 5b,c	RegFactor Analysis (Alpha/Beta)	scMagnify-figures/Fig5/02_regfactor_analysis_Alpha.ipynb
Fig 6a	Intracellular CCI	scMagnify-figures/Fig6/02_intracellular_cci.ipynb
Fig 6b	Comm. Module Analysis	scMagnify-figures/Fig6/test-intracellular_comm_cci.ipynb

Data availability

All data objects (.h5ad, .h5mu) used for the analyses are available on Zenodo: https://doi.org/10.5281/zenodo.YOUR_ZENODO_ID.

Citation

If you use this data or code, please cite:

Decomposing multi-scale dynamic regulatory from single-cell multiomics with scMagnify. Chen*, ..., et al., bioRxiv 2025; doi: https://doi.org/10.1101/YOUR_BIORXIV_ID