Reproducibility files

Figure reconstruction: Python 3 with NumPy and Matplotlib.
Run: python figures.py . figures_output
This uses frozen genuine figure source values. No model fitting or clinical reconstruction occurs in this plotting step.

Stage14 launcher: complete Linux rMATS count-model refitting, with the original compressed input and documented Linux environment. Stage15-17 launchers reproduce their frozen analysis; they additionally need Python, Node and @oai/artifact-tool. See each launcher for dependencies. The analysis backends and figure code were actually run. R entry points were not executed in this environment.

Reassemble original count input: extract counts_part_1.bin and counts_part_2.bin from Additional files4 and5 into one directory, put reassemble_counts.py there, then run python reassemble_counts.py. SHA256 is verified before the output is accepted. No raw FASTQ or BAM is included.

Data-workbook regeneration: Node with @oai/artifact-tool; run node workbook.mjs . output_dir. Set CODEX_PRIMARY_RUNTIME_NODE_MODULES to the installed runtime modules path.

2026-10-27 figure revision: numerical source inputs are unchanged. Figure 2 group means are calculated directly from the shown JC PSI matrix. Figure 4 donor enlargement uses a separate labelled scale. No image-generation model was used. The updated figures.py was executed, with empirical AUC and complete DGLUCY protein comparisons checked.

Journal-format revision, 2026-10-27: figures.py now exports embedded-font vector PDFs at 170 mm width and matching 600 dpi PNG previews. Additional rendering dependencies: pypdf and PyMuPDF (fitz). Figure 4 includes an in-figure key for omitted-sample group; all numerical inputs remain unchanged.
