A repository to calculate genetic correlations with brain imaging phenotypes hosted at https://open.win.ox.ac.uk/ukbiobank/big40/.
It uses snakemake to handle basic dependencies and workflow, and is configured for use with the slurm scheduler.
Clone the repository. In the base directory, run install.sh.
bash install.sh
Install snakemake
conda create -n snakemake -c conda-forge mamba && conda activate snakemake && mamba install -c bioconda -c conda-forge snakemake
Make sure to set the name of your snakemake environment in the the script run_cluster.sh if you want to submit your jobs with slurm.
Configure your input files, see config/input_sumstats.tsv. If your input files have columns recognized by ldsc/munge_sumstats.py, the columns PVAL_COLUMN, A1_COLUMN and A2_COLUMN can be filled with placeholders ("."). You will need to define a sample-size N for every set of summary statistics. Variant-specific sample sizes are not yet supported.
A second set of stumstats can be defined by adding a line to config/config.yaml specifying a path to a second table of summary statistics (e.g., other_sumstats: other_sumstats.tsv). The other_sumstats.tsv table only has three columns ID,N,PATH where N can be . if there is a per-variant sample size column in the summary statistics file. The pipeline can compute genetic correlations between the sumstats in input_sumstats.tsv and other_sumstats.tsv, as well as perform S-LDSC for both.
Define which brain phenotypes you want to compute genetic correlations with by listing their identifiers in the file config/big40_phenotypes.txt. The default is to compare against all 3000+ phenotypes. A list of all phenotypes and their identifiers is contained in resources/big40_metadata.tsv.gz and online at https://open.win.ox.ac.uk/ukbiobank/big40/BIG40-IDPs_v4/IDPs.html/ (December 2021).
Trigger all brain-phentoype genetic correlation steps by running
# check what whill be executed
bash run_cluster.sh --dry-run --quiet all
sbatch run_cluster.sh all
To follow pipeline progress:
tail -f snakemake*log
To run cell-type group analyses with pre-computed annotations provided by the LDSC authors, you can choose one of the available cell type group sets, and request to run the analysis with the following command:
run_cluster.sh results/ldsc_cts/{id}/{ldcts_name}.ok
Where {id} should be substituted with the name of the ID defined in config/input_sumstats.tsv and {ldcts_name} is the name of the set of annotations you would like to estimate enrichments for. Available annotation sets are Multi_tissue_gene_expr, Multi_tissue_chromatin, GTEx_brain, Cahoy, ImmGen, or Corces_ATAC. Enrichments are calculated controling for the annotations in the "baseline ld" model using the HapMap3 variants, as defined by the ldsc authors. Please consider their documentation and accompanying paper for details.
💩 Currently the rules for S-LDSC enrichment with pre-defined sets of varaints are broken, because the download links for the resources have changed.
UCSC-BED-file based annotations can be used to create ldsc-compatible annotation files for S-LDSC, calculate stratified LD-scores, and run enrichment analyses.
BED files need to be placed into folders with the pattern annot_bed/{annot_group}/{annot_id}.bed, where {annot_group} and {annot_id} should be substituted for unique identifiers.
To run S-LDSC with all BED-file annotations inside annot_bed/{annot_group}/, run:
run_cluster.sh results/annot/{annot_group}/all.ok
After that, you can define your own .ldcts-file listing the annotations that you want to test, and run the analysis with either:
# for all sumstats in the main sumstats table:
results/ldsc_cts/{cts_name}.all_ok
# for all sumstats in the "other" sumstats table:
bash run results/ldsc_cts_other/{cts_name}.all_ok