SDPRX is a statistical method for cross-population prediction of complex traits. It integrates GWAS summary statistics and LD matrices from two populations (EUR and non-EUR) to compuate polygenic risk scores.
You can download SDPRX by simply running
git clone https://github.com/eldronzhou/SDPRX.git
SDPRX is developed under python 2.7 but should be compatible with python 3. We recommend you to run SDPRX in the Anaconda so that libraries like numpy and scipy would be pre-installed. If running in the Anaconda environment, the only requirement to run SDPRX would be installing joblib.
The reference LD matrices based on 1000 Genome Hapmap3 SNPs can be downloaded from the following link.
| Populations | Number of SNPs | Size | Link |
|---|---|---|---|
| EUR_EAS | 873,166 | 6.3G | link |
| EUR_AFR | 903,499 | 8.7G | link |
| EUR_SAS | To be uploaded | ||
| EUR_AMR | To be uploaded |
You can use the following command if you wish to use your LD panel to estimate LD matrix.
python calc_ref.py --ref_path1 path_to_plink1_prefix_pop1 --ref_path2 path_to_plink1_prefix_pop2 --chrom chr_number --threads 3 --out ./
The EUR/nonEUR summary statistics should have at least following columns with the same name, where SNP is the marker name, A1 is the effect allele, A2 is the alternative allele, Z is the Z score for the association statistics, and N is the sample size.
SNP A1 A2 Z N
rs737657 A G -2.044 252156
rs7086391 T C -2.257 248425
rs1983865 T C 3.652 253135
...
An example command to run the test data is
python SDPRX.py --ss1 test/EUR.txt --ss2 test/EAS.txt --N1 40000 --N2 40000 --force_shared TRUE --load_ld test/ --valid test/test.bim --chr 1 --rho 0.8 --out test/res_1
A full list of options can be obtained by running python SDPRX.py -h. Below are the required options.
- ss1 (required): Path to the EUR summary statistics.
- ss2 (required): Path to the non-EUR summary statistics.
- N1 (required): Sample size of the EUR summary statistics.
- N2 (required): Sample size of the non-EUR summary statistics.
- load_ld (required): Path to the referecence LD directory.
- valid (required): Path to the bim file for the testing dataset, including the .bim suffix.
- chr (required): Chromosome.
- out (required): Path to the output file containing estimated effect sizes.
- rho (required): Trans-ethnic genetic correlation output by PopCorn between 0 and 1. Default is 0.8.
- force_shared (required): Whether to force sharing of effect sizes between populations. Default is True.
- n_threads (optional): number of threads to use. Default is 1.
For real data analysis, it is recommended to run each SDPRX on each chromosome in parallel, and using 3 threads for each chromsome.
There are two output files corresponding to the adjusted effect sizes for EUR (e.g. res_22_1.txt) and non-EUR population (e.g. res_22_2.txt). One can use PLINK to derive the PRS.
plink --bfile test_geno --score res_22_1.txt 1 2 3 header --out test_1 # EUR
plink --bfile test_geno --score res_22_2.txt 1 2 3 header --out test_2 # non-EUR
If a validation dataset is available, one can further learn a linear combination of PRS (a * PRS_1 + (1-a) * PRS_2 for a grid of a ranging from 0 to 1 by a step of 0.5) based on the best performance in the validation dataset.
The weights for real traits analyzed in our paper can be downloaded here. The summary information in the format of PGS catalog can be found here.
Zhou G, Chen T, Zhao H. SDPRX: A statistical method for cross-population prediction of complex traits. Am J Hum Genet. 2023 Jan 5;110(1):13-22. doi: 10.1016/j.ajhg.2022.11.007.