w# LDSC

Wrapper pipeline based on https://github.com/bulik/ldsc.

Most of the work is done by the wdl itself, but some preprocessing steps are needed, mainly due to the fact that the nature of the input sumstats can be different.

WDL

The wdl takes a list of sumstats and generates heritabilities and (optional) genetic correlation between all N(N-1)/2 pairs or if two separate lists are passed then only between cross N*M pairs.

Inputs

Global parameters

Parameter	Default	Description
ldsc_rg.meta_fg	—	Metadata table for primary summary statistics (TSV).
ldsc_rg.meta_other	—	Metadata table for secondary sumstats. Pass the same file as meta_fg for a self-comparison.
ldsc_rg.name	—	Output prefix for result files.
ldsc_rg.only_het	—	If true, computes only heritabilities (not genetic correlations).
ldsc_rg.population	—	Population label used to resolve the LD score file (e.g., FIN, EUR).
ldsc_rg.docker	eu.gcr.io/finngen-sandbox-v3-containers/ldsc:rsid_munge	Docker image used for all pipeline tasks.
ldsc_rg.ld_root	gs://finngen-production-library-green/ldsc/POP_ld.txt	GCS path template for LD score file lists; POP is replaced by population.
ldsc_rg.snplist	gs://finngen-production-library-green/ldsc/w_hm3.snplist	SNP list file for LD score regression.
ldsc_rg.filter_chunk_size	30	Number of phenotypes per premunge/munge scatter shard. Use smaller values for testing.
ldsc_rg.couples_chunk_size	500	Number of pairs per multi_rg scatter shard. Use smaller values for testing.
ldsc_rg.multi_rg.cpus	—	Number of CPUs per multi_rg shard.
ldsc_rg.munge_fg.args	—	(Optional) Extra arguments passed to ldsc.py for the fg munge step.
ldsc_rg.munge_other.args	—	(Optional) Extra arguments passed to ldsc.py for the other munge step.
ldsc_rg.multi_rg.args	—	(Optional) Extra arguments passed to ldsc.py for the rg step.

Premunge parameters — primary list (meta_fg)

Parameter	Description
ldsc_rg.premunge_fg.beta_col	Column name for effect size (beta).
ldsc_rg.premunge_fg.p_col	Column name for p-value.
ldsc_rg.premunge_fg.a1_effect_col	Column name for effect allele.
ldsc_rg.premunge_fg.a2_ne_col	Column name for non-effect allele.
ldsc_rg.premunge_fg.rsid_col	(Optional) Column name for rsIDs.
ldsc_rg.premunge_fg.chrom_col	(Optional) Column name for chromosome (required if no rsID column).
ldsc_rg.premunge_fg.pos_col	(Optional) Column name for position (required if no rsID column).

Premunge parameters — secondary list (meta_other)

Only required when meta_other differs from meta_fg. The parameters mirror those above:

Parameter	Description
ldsc_rg.premunge_other.beta_col	Column name for effect size (beta).
ldsc_rg.premunge_other.p_col	Column name for p-value.
...	...

The metadata tables should be structured as PHENO\tPATH\tN where N is the total number of valid cases+controls of each pheno.

C3_BREAST_EXALLC	gs://fg-cromwell_fresh/munge_fg/d17c3b71-2510-4d89-8bfb-3f788b50bd59/call-munge/shard-0/C3_BREAST_EXALLC.premunge.gz	110611
C3_BRONCHUS_LUNG_EXALLC	gs://fg-cromwell_fresh/munge_fg/d17c3b71-2510-4d89-8bfb-3f788b50bd59/call-munge/shard-1/C3_BRONCHUS_LUNG_EXALLC.premunge.gz	180418
C3_PROSTATE_EXALLC	gs://fg-cromwell_fresh/munge_fg/d17c3b71-2510-4d89-8bfb-3f788b50bd59/call-munge/shard-2/C3_PROSTATE_EXALLC.premunge.gz	83146
G6_PARKINSON	gs://fg-cromwell_fresh/munge_fg/d17c3b71-2510-4d89-8bfb-3f788b50bd59/call-munge/shard-3/G6_PARKINSON.premunge.gz	224566
H7_AMD	gs://fg-cromwell_fresh/munge_fg/d17c3b71-2510-4d89-8bfb-3f788b50bd59/call-munge/shard-4/H7_AMD.premunge.gz	214660

Munging

The wdl now contains internally a premunge_ss step where input sumstats are processed to match the LDSC notation, which is

SNP	A1	A2	BETA	P
rs74337086	A	G	0.0923	0.5059
rs76388980	A	G	0.1227	0.2945
rs562172865	T	C	-0.0262	0.8142
rs780596509	A	G	-0.2202	0.1545
rs778009914	A	G	-0.3938	0.3044
rs564223368	T	C	0.2195	0.03913
rs71628921	C	A	0.1763	0.3682
rs577189614	A	G	0.0845	0.5341
rs77357188	T	C	-0.0414	0.3383

Therefore now the the inputs also require to pass the relevant column names for the munging. In case the data is not in rsid format, the script will automatically map chrom/pos --> rsid if needed. chrom_col and pos_col are required only if the rsid_col is missing

Long description

A brief summary of the logic of the wdl.

only_het if set to true only produces heritabilities and does not compute genetic correlations. Please make sure it's your intention to compute correlations before setting it to false.

Both meta_fg and meta_other are always required. Pass the same file for both to run a self-comparison (all N(N-1)/2 pairs within the list). Pass two different files to run a cross-comparison (N*M pairs). The growth is quadratic so it is recommended to test with a smaller set first.

filter_meta splits each input list into chunks of filter_chunk_size phenotypes. The two lists are chunked independently. If the files are identical the second scatter is skipped entirely. This step is fast and its output is cached.

Each chunk is passed in parallel to premunge_ss (premunge_fg for meta_fg, premunge_other for meta_other) which prepares the sumstats for ldsc as described above. Column name inputs are specified per call, allowing the two lists to have different formats.

Each premunged chunk is passed to munge_ldsc which runs ldsc munging and computes per-phenotype heritability. The heritability outputs from both scatter arms are combined and passed to gather_h2, which builds a summary table, JSON, and merged logs.

return_couples builds all unique phenotype pairs across the two lists, splits them into chunks of couples_chunk_size pairs each, and for each chunk produces the subset of munged sumstat paths required — so only the necessary files are localized per shard.

Each chunk of pairs is passed to multi_rg where a wrapper script runs ldsc.py --rg in parallel. Increasing CPUs via ldsc_rg.multi_rg.cpus speeds up each shard.

Finally gather_summaries merges all multi_rg outputs into a single table and log:

p1	p2	rg	se	z	p	h2_obs	h2_obs_se	h2_int	h2_int_se	gcov_int	gcov_int_se
AB1_AMOEBIASIS	AB1_AMOEBIASIS	1.0006	0.0009	1139.1313	0.0	0.0009	0.0013	0.9682	0.0061	0.9682	0.0061
AB1_AMOEBIASIS	AB1_ANOGENITAL_HERPES_SIMPLEX	0.5613	0.7023	0.7992	0.4242	0.0039	0.0015	0.9872	0.0069	0.0007	0.0044
AB1_AMOEBIASIS	AB1_ARTHROPOD	-1.0562	1.0868	-0.9718	0.3311	0.0024	0.0015	0.9977	0.0071	0.0089	0.005
AB1_AMOEBIASIS	AB1_ASPERGILLOSIS	0.7488	0.9701	0.7719	0.4402	0.0019	0.0016	0.981	0.0065	-0.0074	0.0048

LD scores

The LD score file is resolved from ld_root by substituting POP with the value of population. Set population to fin or eur to use the prebuilt FinnGen or 1000 Genomes European LD scores respectively.

ldsc_rg.multi_rg.args is an optional input for passing extra flags directly to ldsc.py.