Analysis code and figure/table outputs for the study, from the Kundishora Lab (Hale & Keller et al.).
Brain arteriovenous malformations (bAVMs) are high-flow cerebrovascular lesions associated with a lifelong risk of intracranial hemorrhage, yet no biologically anchored framework exists to predict their clinical trajectory or guide medical therapy. Although activating somatic variants in KRAS and BRAF drive sporadic bAVMs, the relationship between genotype, disease severity, and clinical behavior has remained poorly defined, and molecular diagnosis has largely been restricted to surgically resected tissue.
This study assembles the largest integrated genomic–phenomic cohort of sporadic bAVMs to date (n = 475) — harmonizing somatic genotypes, clinical phenotypes, angioarchitectural features, and neuroanatomical localization — to define a genotype-stratified natural history of the disease. Pathogenic somatic variants were identified in 67.4% of lesions, dominated by KRASG12D; variant-positive lesions presented and ruptured roughly two decades earlier than panel-negative bAVMs, variant allele fraction (VAF) tracked inversely with age at rupture as a dose-dependent modifier of severity, and endovascular-enabled liquid biopsy detected KRASG12D in perinidal cell-free DNA — proof-of-concept for in situ molecular diagnosis without surgical resection. Together these results outline a framework for precision diagnosis, biologically anchored prognosis, and genotype-directed therapeutic trials in sporadic bAVMs.
This repository holds the analysis code and the figure/table/statistics outputs underlying those findings.
analysis/
00_data_prep/ 00 mask PHI · 01 clean the cohort master → analysis-ready dataset
01_main_analysis/ producers + composers, named NN_<TARGET>_<desc>.R and grouped so
each main figure sits with its companion Extended Data figures
(F1–F4, ED1–ED10, T1 = Table 1, ST = Supplementary Tables)
pipeline/ run_all.R orchestrator + figure-assembly / stats-manifest helpers
helper_scripts/ shared R utilities (palettes, themes, plotting, stats)
results/
figures/ manuscript-facing deliverable, organized by figure + panel:
Figure 1/ … Figure 4/ Fig N (composite) + Fig NA, NB, …
Extended Data Fig 1/ … 10/ ED Fig N (composite) + ED Fig NA, …
Tables/ Table 1 + Supplementary Tables
stats/ text readouts of the reported statistics
_sessionInfo.txt R session the outputs were generated under
No data are committed to this repository — see Data availability below.
analysis/01_main_analysis/SCRIPT_INDEX.md maps every script to the figure(s),
table(s), and statistics it produces.
Everything runs as one sequence from 00 to the end:
# R 4.5.1
source("analysis/pipeline/builders/run_all.R")run_all.R (1) reprocesses the cohort data, (2) runs every producer in
analysis/01_main_analysis/ in number order — each writing its panels, tables, and a
statistics fragment, (3) aggregates the fragments into
results/stats/manuscript_stats.rds, (4) validates the manifest and panel-token
uniqueness, and (5) organizes the panels into the labeled results/figures/ tree.
Figure numbers and panel letters come from the frozen
results/stats/panel_assignments.rds.
No data are committed to this repository. The cleaned cohort dataset that drives the
analysis is available from the corresponding author on reasonable request, subject to
the governing IRB and data-use agreements. With that dataset placed under data/,
analysis/00_data_prep/01_clean_master.R and the full producer chain regenerate every
figure, table, and statistic.
One extended-data panel (dPCR validation) draws its source measurements from the companion variant-detection dataset; its rendered panel is included here.
If you use this code, please cite the associated manuscript (Hale & Keller et al.).