s-ldsc

S-LDSC

Requirements

Install dependencies with uv:

uv venv --python 3.13
source .venv/bin/activate
uv pip install -r requirements.txt

Additionally, conda is required to run the ldsc step.

Optional (recommended): download the processed data (9.1G, feel free to only download a subset):

• S-LDSC reference files, originally in (https://zenodo.org/records/10515792) but packaged here as well for your convenience.
• Models
• Traits
• Results (models x traits)

mkdir -p results
uv run hf download songlab/ldsc --repo-type dataset --local-dir results/
cd results
tar -xf output.tar

Running

Take a look at workflow/Snakefile rule all for example targets. The first one will run S-LDSC on one model on one trait.

uv run snakemake --cores all --use-conda

# Snakemake sometimes gets confused about which files it needs to rerun and this forces
# not to rerun any existing file
uv run snakemake --cores all --touch
# to output an execution plan
uv run snakemake --cores all --dry-run

LDSC jobs (e.g. running model X on trait Y) is by default parallelized as 1 job per core. This works as long as you have enough memory (e.g. when asking for a complete savio3_htc or savio4_htc node). When running on a node with less memory, should reduce the parallelization of the ld_score and run_ldsc_annot rules in workflow/rules/ldsc.smk.

To add a new model {model}, place a parquet file with column score in either of these locations:

• results/variant_scores/{model}.parquet

  Variants should be in the standard S-LDSC order (e.g. see the 9,997,231 variants in https://huggingface.co/datasets/songlab/ldsc)

• results/features/{model}.parquet

  This corresponds to the variants above with pos != -1. pos == -1 are a tiny fraction where we were not able to liftover from hg19 to hg38. This complexity is because S-LDSC uses hg19 but we do most our work with newer annotations in hg38.

To add a new trait {trait}, place it under results/sumstats_107/{trait}.sumstats.gz (please check format of other traits).