LEAF-1 figure reproducibility

Code and processed data to regenerate the figures in the LEAF-1 manuscript. Each notebook reads tables staged in the repository and writes figure files. Model checkpoints, H5 files, and raw sequencing data are not required.

Organization

leaf-1-figures/
├── cfdna/
│   ├── notebook/   one notebook per figure (see map below)
│   ├── data/       input tables read by the notebooks
│   └── output/     generated figures (.png, .pdf) and derived tables
├── scatac/
│   ├── notebook/   one notebook per figure (see map below)
│   ├── data/       input tables read by the notebooks
│   └── output/     generated figures (.png, .pdf) and derived tables
├── other/
│   ├── notebook/   model-overview pretraining curves
│   ├── data/       input tables read by the notebook
│   ├── diagrams/   schematic source files
│   └── output/     generated figures and derived tables
├── src/leaf1figures/        shared matplotlib style (set_paper_style, savefig, OKABE_ITO)
├── NOTEBOOK_PROVENANCE.tsv  notebook, figure, source analysis notebook, input tables
├── DATA_MANIFEST.tsv        size and SHA-256 for each input table
├── FIGURE_OUTPUTS.tsv       size and SHA-256 for each output
├── environment.yml
├── CITATION.cff
├── LICENSE
└── LICENSE-DATA

Notebooks, input tables, and outputs are grouped by modality (cfdna, scatac). Each notebook reads only from the data/ directory next to it and writes only to the adjacent output/ directory. Notebooks are committed with their cell outputs.

Figure map

scATAC (scatac/notebook/):

Notebook	Manuscript figure
scatac_celltype_classification.ipynb	Figure 2b-f
scatac_tf_motif_accessibility.ipynb	Figure 3a-d
scatac_fragment_umap.ipynb	Figure 3e-f
supp_scatac_classification_extended.ipynb	Supplementary Figures 1-3
supp_scatac_embedding_vs_binning_resolution.ipynb	Supplementary Figure 4
supp_scatac_segment_ablation.ipynb	Supplementary Figure 5

cfDNA (cfdna/notebook/):

Notebook	Manuscript figure
cfdna_representation_benchmark.ipynb	Figure 4b-e
cfdna_mil_detection.ipynb	Figure 5b-e
cfdna_ctdna_detection.ipynb	Figure 5f-g
cfdna_ichorcna_genomewide.ipynb	Figure 6b
kidney_zeroshot.ipynb	Figure 6c-d
kidney_zeroshot_leaf1_vs_binning_auc.ipynb	Figure 6e
supp_cfdna_per_cancer_metrics.ipynb	Supplementary Figure 8
supp_cfdna_subsampling.ipynb	Supplementary Figures 7, 9-13

Model overview (other/notebook/):

Notebook	Manuscript figure
pretraining_curves.ipynb	Pretraining curves (model overview)

Setup

conda env create -f environment.yml
conda activate leaf1-paper-repro

Running

Open a notebook in Jupyter and run it top to bottom. The first cell resolves the data/, output/, and src/ paths, so a notebook runs from its own directory, from its modality directory, or from the repository root.

To regenerate all figures without opening Jupyter:

jupyter nbconvert --to notebook --execute --inplace */notebook/*.ipynb

PNG and table outputs are deterministic under fixed random seeds. PDF outputs are visually identical across runs but not byte-identical, because matplotlib records a creation timestamp in PDF metadata. The checksums in FIGURE_OUTPUTS.tsv correspond to the committed files.

Updating a figure

1. Identify the notebook from the figure map.
2. To change styling or layout, edit the notebook. To change the underlying numbers, replace or add a table in the modality data/ directory and update the notebook to read it.
3. Re-run the notebook so output/ and the embedded cell outputs refresh.
4. If data/ or output/ files changed, update DATA_MANIFEST.tsv and FIGURE_OUTPUTS.tsv accordingly.

License

• Code (notebooks and src/leaf1figures/): MIT, see LICENSE.
• Data and figures (*/data/, */output/): CC-BY-4.0, see LICENSE-DATA.
• Some bundled data are third party and retain their original licenses (see below).

Please cite the LEAF-1 paper (CITATION.cff).

Data sources

• Transcription-factor motifs (scATAC motif figure): JASPAR database (CC0).
• scATAC-seq metadata and atlas-derived tables: human single-cell chromatin-accessibility atlas of Zhang et al., 2021, Cell 184(24):5985-6001.e19, doi:10.1016/j.cell.2021.10.024, source data at http://catlas.org/. Sample identifiers are public consortium codes.
• cfDNA tables: model outputs for the cell-free DNA cohort of Cristiano et al., 2019, Nature 570:385-389. The underlying fragmentation profiles are available through FinaleDB (Zheng et al., 2021, Bioinformatics 37:2502-2503, doi:10.1093/bioinformatics/btaa999; http://finaledb.research.cchmc.org/), a public database of cell-free DNA fragmentation patterns. Please cite Cristiano et al. (2019) and FinaleDB if you reuse these.
• Kidney prediction tables: model outputs for an independently profiled, de-identified clinical cohort (pseudonymous codes only; no names, dates, or other identifiers).