LDSC (LD Score Regression): Commands and Limits

LDSC (LD Score regression) is a free command-line program that estimates SNP heritability and genetic correlation from genome-wide association study (GWAS) summary statistics, and that also computes LD Scores. This page is for researchers and research-support staff who need to know what the tool does, which commands and flags map to which analysis, where its documented limits are, and which paper to cite for each use. Details below were read from the project’s GitHub README and wiki on 2026-11-07.

What LDSC (LD Score regression) is

The project README describes ldsc as “a command line tool for estimating heritability and genetic correlation from GWAS summary statistics” that “also computes LD Scores.” The current release listed there is v1.0.1, and the code is licensed under the GNU GPL v3. It is a Python program run from a terminal, not a web service. The acronym “LDSC” is also used for unrelated things, so this page uses “LD Score regression” to name the method.

The wiki tutorials group the tool into four basic functions: estimating LD Scores, estimating heritability (h2) and partitioned heritability, estimating genetic correlation, and using the LD Score regression intercept.

Who maintains it and what it covers

The repository at github.com/bulik/ldsc lists Brendan Bulik-Sullivan (Broad Institute) and Hilary Finucane (MIT Mathematics) as authors. Support routes named in the README are the project wiki, an FAQ page and the ldsc_users Google group. A README banner dated January 16, 2026 points to a newer Python 3 implementation at github.com/CBIIT/ldsc, with a web browser for LD Scores at ldlink.nih.gov/ldscore.

Pre-computed reference files are hosted at data.broadinstitute.org/alkesgroup/LDSCORE. The README lists European and East Asian LD Scores computed from 1000 Genomes, plus partitioned LD Scores.

How to run it

These steps follow the README install notes and the wiki page “Heritability and Genetic Correlation.”

Install. Clone the repository (git clone https://github.com/bulik/ldsc.git), change into the folder, create the environment from environment.yml with Anaconda (conda env create --file environment.yml), then activate it (source activate ldsc).
Test. Run ./ldsc.py -h and ./munge_sumstats.py -h; both should print help text.
Prepare summary statistics. Run munge_sumstats.py with --sumstats, --N, --out and --merge-alleles w_hm3.snplist. The input needs a per-SNP identifier, two alleles, a sample size, a P-value and a signed statistic. Defaults filter to INFO > 0.9, MAF > 0.01 and 0 < P ≤ 1, and drop non-SNPs, strand-ambiguous SNPs and duplicate rs numbers. Output is <out>.sumstats.gz and a .log file.
Estimate heritability. Run ldsc.py --h2 x.sumstats.gz --ref-ld-chr eur_w_ld_chr/ --w-ld-chr eur_w_ld_chr/ --out x_h2.
Estimate genetic correlation. Run ldsc.py --rg a.sumstats.gz,b.sumstats.gz with the same reference options and an --out prefix. With three files, --rg a,b,c reports rg(a,b) and rg(a,c).

Flag notes from the wiki: in --ref-ld-chr the value is a file prefix without .l2.ldscore, and “@” marks the chromosome position. For non-partitioned analyses --w-ld-chr uses the same files as the reference. --samp-prev and --pop-prev convert heritability to the liability scale (the result is nan for quantitative traits) and change h2 but not the intercept.

Strengths and limits

Documented strengths: it works from summary statistics rather than individual-level genotypes, it has ready-made reference LD Scores, and one tool covers heritability, partitioned heritability, genetic correlation and the intercept (the 2015 paper’s title frames the method as distinguishing confounding from polygenicity). The limits below come from the wiki.

Situation	What the wiki says
Population mismatch	The provided LD Scores are from 1000 Genomes European samples; other populations need matched LD Scores.
Few SNPs	Fewer than 200,000 remaining SNPs is usually bad; ldsc warns.
Low signal	munge_sumstats.py warns if the mean chi-square is below 1.02.
GC-corrected input	Genomic-control-corrected data bias h2 and genetic covariance downward, but not genetic correlation.
Overlapping samples with `--no-intercept`	Constraining intercepts (h2 intercepts to 1, cross-trait to 0) can mislead and can yield rg outside [-1, 1].
`--h2` versus `--rg`	Estimates can differ slightly because rg uses only SNPs shared by both traits.

Alternatives and related tools

Tool	Relationship named in the README
CBIIT/ldsc	Newer Python 3 implementation announced in the README banner (January 16, 2026).
LDlink LD Score browser	Web browser for LD Scores linked from the same banner.
LD Hub	Web resource cited in the README: Zheng et al., Bioinformatics 2016 (doi 10.1093/bioinformatics/btw613).

For finding the GWAS that supply summary statistics, see the GWAS Catalog guide. For reporting genetic-instrument studies, see STROBE-MR.

Which paper to cite

Use	Citation named in the README
Software and intercept	Bulik-Sullivan et al., LD Score regression distinguishes confounding from polygenicity in genome-wide association studies, Nature Genetics 2015 (PMID 25642630)
Genetic correlation	Bulik-Sullivan et al., An atlas of genetic correlations across human diseases and traits, Nature Genetics 2015 (ng.3406)
Partitioned heritability	Finucane et al., Nature Genetics 2015 (ng.3404)
Continuous annotations	Gazal et al., Nature Genetics 2017 (ng.3954)

Also record the software version and the reference LD Score files you used; see FAIR4RS software citation principles and computational reproducibility.

FAQ

What does LDSC stand for in genetics?

In genetics, LDSC refers to the LD Score regression software (ldsc) and the method behind it. The README describes it as a tool for estimating heritability and genetic correlation from GWAS summary statistics.

Is LDSC free to use?

Yes. The repository lists the GNU GPL v3 license.

Does LDSC need individual-level genotype data?

No. Its inputs are GWAS summary statistics, processed with munge_sumstats.py, plus reference LD Scores.

Can I use it on non-European cohorts?

The wiki states the provided LD Scores are European and that other populations need matched LD Scores. The README also lists East Asian LD Scores among the hosted files.

Where do I get help?

The README names the wiki, the FAQ and the ldsc_users Google group.

Sources

ldsc README, github.com/bulik/ldsc, read 2026-11-07
ldsc wiki: Heritability and Genetic Correlation, read 2026-11-07
PubMed 25642630, record checked 2026-11-07

Ask about LDSC (LD Score Regression): Commands, Limits and Citations

Ask your first 2 questions free below. Subscribers get 150 a day for $29 a month.

An AI assistant specialized in research administration. It cites the sources behind every answer, labels web answers and says when it can't answer.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays free to read, with no account and no card.