RSparsePro is a command line tool for robust fine-mapping in the presence of LD mismatch. For analysis conducted in the paper, please refer to RSparsePro_LD_analysis.

RSparsePro takes GWAS summary statistics (z-scores) and an LD reference panel (LD matrix) as input. In RSparsePro, we introduce a latent variable to address potential inconsistencies between the GWAS population and the LD reference panel, enabling robust fine-mapping.

Figure 1: RSparsePro overview.

SharePro was developed under Python 3.9.7 but should be compatible with other versions of Python 3. The following Python modules are required:

• numpy
• scipy
• pandas

To install RSparsePro for robust fine-mapping:

git clone https://github.com/zhwm/RSparsePro_LD.git
cd RSparsePro_LD
pip install -r requirements.txt 

To test the installation and display basic usages:

python src/rsparsepro_ld.py -h

Example input files are included in the dat directory.

1. The z-score file: The column for z-scores must have the header 'Z', and the column for SNP index must have the header 'RSID'. Both columns are required. You may include additional columns in the file, but RSparsePro will not utilize other information. An example can be found at test.z.txt.

2. The LD file: The LD file contains the LD matrix, which includes the Pearson correlation between each pair of variants, with values separated by a single space. An example can be found at dat/ld.txt.

While RSparsePro can handle allele flipping, it is strongly recommended to ensure the REF/ALT alleles used in calculating the LD matrix match those from the GWAS study.

We use --z and --ld to indicate the path to the zscore file and the ld file, respectively. Additionally, we can specify the path to save results with --save.

python src/rsparsepro_ld.py \
--z dat/test.z.txt \
--ld dat/ld.txt \
--save dat/test

After a successful execution of RSparsePro, we will append three new columns to the input z-score file in the output result file: PIP for posterior inclusion probability, z_estimated for estimated latent z-scores, and cs for credible sets. Variants in credible sets will have non-zero values in the cs column.

Additionally, the log file records the execution process of the algorithm.

If you find RSparsePro useful, please cite:

Wenmin Zhang, Tianyuan Lu, Rob Sladek, Josée Dupuis, Guillaume Lettre. Robust fine-mapping in the presence of linkage disequilibrium mismatch