Local genetic architecture of Restless Legs Syndrome

Analysis code for the manuscript "Local Genetic Architecture of Restless Legs Syndrome and Parkinson's Disease Across Sleep, Psychiatric, and Movement Disorders."

The pipeline is anchored on Restless Legs Syndrome (RLS) and screens a panel of sleep, psychiatric, and movement traits, including Parkinson's disease (PD), through four layers:

1. Global genetic correlation (LDSC)
2. Local genetic correlation (LAVA, per ~2,495 LD blocks)
3. Colocalization (coloc.abf at every block passing the local screen, plus prior/window sensitivity and coloc.susie at MEIS1)
4. Cross-ancestry comparison of the TOX3 RLS-PD signal (EUR / AFR / AMR)

Summary-statistic files and LD reference panels are not redistributed here (they are public or collaborator-provided; see Data). The scripts document every input and its source.

Running the pipeline

After setting the paths (see Configuration), the top-level driver runs every stage in order:

bash run_all.sh

run_all.sh is the single source of truth for execution order. In brief:

1. Setup - build the toolchain; download LD references (1000G EUR/AFR/AMR, HapMap3).
2. Munge - download and harmonize every GWAS (RLS, PD, panel).
3. Global rg - LDSC: RLS x panel and PD x panel.
4. Local rg - LAVA genome-wide, then cross-ancestry LAVA at TOX3.
5. Coloc - systematic coloc.abf screen, targeted loci, then the robustness analyses (prior, window, coloc-SuSiE).
6. Cross-ancestry - TOX3 forest + per-SNP (beta-beta) concordance.
7. Figures.

Within a stage the scripts are largely independent, with two ordering rules that run_all.sh already encodes: each annotate_* script must run before the munge_* scripts that consume its rsID-annotated output, and the coloc / figure steps consume the LAVA output. These are the scripts as run; the RLS-anchored steps require the access-controlled multi-ancestry RLS data (see Reproducibility).

Repository layout

Scripts are grouped by pipeline stage; run the stages in order (or use run_all.sh).

scripts/1_setup/ - toolchain and LD references

• 00_setup_environment.sh - build the version-pinned LDSC / LAVA / PLINK toolchain (records the LDSC Python-3 patches)
• 01b_get_ldsc_ref.sh - LDSC LD scores + HapMap3 weights (1000G Phase 3)
• 01c_smoketest.sh - end-to-end LDSC validation on synthetic data
• 04_get_1000g_eur.sh - 1000G Phase 3 EUR PLINK panel (LAVA / clumping)
• dl_g1000_panels.sh - 1000G AFR + AMR panels (cross-ancestry LAVA)
• build_hm3_hg38_map.sh - HapMap3 hg38->rsID map (to annotate hg38 PD sumstats)

scripts/2_munge/ - download and harmonize all GWAS

• RLS: 16_munge_akcimen.sh (Akcimen EUR), dl_rls_xanc.sh + munge_rls_xanc.sh (multi-ancestry AFR/AMR)
• PD: pd_01_download.sh + pd_02_munge.sh (Nalls + FinnGen), munge_pd_rizig.sh (AFR), munge_pd_xanc.sh + annotate_pd_xanc.py (GP2 EUR / Loesch AMR), munge_pd_largepd.sh (LARGE-PD AMR)
• Panel: annotate_sleep.sh (rsID annotation), munge_panel.sh (psychiatric + sleep), munge_insomnia.sh, munge_sleep2.sh (OSA + sleep duration), munge_sleep3.sh (sleep-duration rsID fix), munge_rg_daytime.sh (daytime sleepiness), munge_narcolepsy_ollila.sh (narcolepsy, Ollila 2023 EUR)

scripts/3_global_rg/ - LDSC genetic correlation

• panel_rg_akcimen.sh (primary: RLS-Akcimen x panel), panel_rg.sh (RLS + PD x panel), insomnia_rg.sh, pd_03_rg.sh (PD x RLS)

scripts/4_local_rg_lava/ - LAVA local genetic correlation

• panel_lava.R (genome-wide bivariate local rg; panel_lava2.R is the parallelizable variant), panel_lava_run.sh / panel_lava2_run.sh (runners)
• pd_05_lava.R (the genome-wide LAVA engine driven by LAVA_PAIRS), pd_xanc_run_lava.sh (cross-ancestry and GP2 scans, incl. TOX3; calls pd_05_lava.R)
• run_narco_lava.R (narcolepsy against RLS and PD, after the switch to Ollila 2023)

scripts/5_coloc/ - colocalization (coloc.abf)

• coloc_screen.R (screen over a subset of trait pairs) and coloc_screen_full.R (the full-panel screen behind results/coloc_screen_full_supplementary.csv), plus targeted runs coloc_rlspd.R (RLS x PD incl. TOX3), coloc_meis1.R, coloc_chr1.R (each with a .sh runner that extracts the region and calls the R script), and run_narco_coloc.R (narcolepsy hits)
• robustness: coloc_prior_sens.R (PP.H4 across colocalization priors), coloc_window_sens.R (PP.H4 across window widths around each lead variant), coloc_susie_meis1.R (coloc-SuSiE at MEIS1, allowing multiple causal variants per trait) - together Supplementary Table S4

scripts/6_cross_ancestry_tox3/ - TOX3/CASC16 cross-ancestry

• run_xanc_coloc_afr.R (African-ancestry local rg at every colocalizing locus; the TOX3 row reproduces the primary estimate and acts as a positive control)
• forest_tox3.py (local-rg forest across datasets and ancestries; Figure 5), betabeta_extract.sh + betabeta_plot.py (per-SNP effect concordance)

scripts/7_figures/

• forest_rg.R (global-rg forest, Figure 2), coloc_selectivity.py (local rg against colocalization posterior, Figure 3)

Both read their numbers from the tables in results/, so a figure cannot drift away from the reported values.

Results

results/ holds the machine-readable numbers behind the manuscript; results/README.md describes each file. In brief: global_genetic_correlation_ldsc.csv (LDSC, Supplementary Table S2), colocalization_hits.csv (Table 1), coloc_screen_full_supplementary.csv (all 74 local correlations carried to colocalization, with Bonferroni and Benjamini-Hochberg columns over the 2,821 bivariate tests, Supplementary Table S3), coloc_prior_sensitivity.tsv and coloc_window_sensitivity.tsv plus susie_meis1_rls_*.csv (Supplementary Table S4), tox3_crossancestry_local_rg.csv (Figure 5), and xanc_coloc_afr.csv (African-ancestry check at the colocalizing loci).

Software

• LDSC v1.0.1 (Python-3 port), conda env: Python 3.9, numpy 1.23.5, scipy 1.9.3, pandas 1.4.4
• LAVA v0.1.5 - R 4.3.3
• coloc v5.2.3 (coloc.abf and coloc.susie) - R 4.3.3
• susieR v0.14.2 - R 4.3.3 (multi-signal fine-mapping behind coloc-SuSiE)
• PLINK v1.9 / v2.0
• LD reference: 1000 Genomes Phase 3 (EUR / AFR / AMR); HapMap3 SNP list for munging

Exact package pins and the LDSC Python-3 patches are recorded in scripts/1_setup/00_setup_environment.sh.

Configuration

Every script refers to the analysis root through the placeholder /path/to/analysis (and the private RLS input through /path/to/akcimen_rls_sumstats). Point them at your own directories before running, for example:

grep -rl /path/to/analysis scripts/ run_all.sh | xargs sed -i "s#/path/to/analysis#$PWD/work#g"

Under the analysis root the scripts expect data/gwas/, data/ldsc_ref/, data/reference/, results/, and tools/ldsc/ (the stage-1 scripts create the reference and tool directories).

Data

All GWAS summary statistics, with releases and accessions, are listed in Supplementary Table S1 of the manuscript. Every dataset is publicly available except the multi-ancestry RLS GWAS, which was provided by Akcimen et al. Raw sumstats and LD panels are omitted here for size and licensing; the download scripts record their sources.

Reproducibility

These are the analysis scripts as run, in a Linux (WSL Ubuntu) environment. The public-data steps can be reproduced after installing the toolchain (stage 1) and repointing the paths; the RLS-anchored steps additionally require the multi-ancestry RLS summary statistics, which are access-controlled through Akcimen et al., so full end-to-end re-execution depends on obtaining that data. Every step, parameter, and data source is documented for transparency.

License

MIT - see LICENSE.