Please note that QuickPRS has now been replaced by GigaPRS, so this page exists only for completeness.
QuickPRS is a tool for constructing PRS from GWAS summary statistics. Its main advantage over MegaPRS is simplicity (in particular, it does not require a reference panel). On the other hand, QuickPRS can only be used with SNP-based summary statistics for human traits (whereas MegaPRS can be used with non-SNP data, and for non-human traits).
QuickPRS uses pre-computed SNP-SNP correlations, constructed using data from the 1000 Genome Project, the UK Biobank or the Haplotype Reference Consortium. We currently provide correlations for ten (sub) populations: AFR (661 African individuals from 1000GP), AMR (347 Admixed American individuals from 1000GP), CARAFR (6578 Carribean and African individuals from UKBB), CEU (5994 European individuals from HRC), CHI (10028 Chinese individuals from HRC), EAS (504 East Asian individuals from 1000GP), FIN (3256 Finnish individuals from HRC), INDPAK (6706 Indian and Pakistani individuals), SAS (489 South Asian individuals from 1000GP) and UK (9880 white British individuals from UKBB). Click here to see a principal component plot of the ten populations.
For each population, we provide correlations based on approximately 1.1M non-ambiguous, autosomal HapMap3 SNPs with MAF>0.01, or based on approximately 2.5M non-ambiguous, autosomal imputed SNPs with MAF>0.01. Please note that you should only use QuickPRS when you have summary statistics for the majority of the SNPs used to generate the correlations. If you are missing summary statistics for more than, say, 20% of the SNPs, then consider switching to MegaPRS.
Always read the screen output, which suggests arguments and estimates memory usage.
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Preparation:
First ensure that your summary statistics are in the format required by LDAK (see Summary Statistics for details). We then suggest you download the following file, and use it to calculate the overlap between your summary statistics and the HapMap3 SNPs.
Details of HapMap3 SNPs (All populations)
Be aware that the SNP names in your summary statistics should be rsIDs (e.g, names such as rs116720794 or rs3131972). If this is not the case, you may be able to use the above file (which contains genomic positions for both the hg19 and hg38 assemblies) to convert the current SNP names to rsIDs.
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Download and extract pre-computed SNP-SNP correlations:
You should select the population that best matches your summary statistics. For example, if your summary statistics come from a GWAS of white Europeans, you should probably use the UK correlations.
Note that when using a UNIX operating system, you can download a file by typing wget followed by its web address (you can obtain the web address of the files below by right-clicking on the corresponding links).
AFR SNP-SNP Correlations (HAPMAP)
AMR SNP-SNP Correlations (HAPMAP)
CARAFR SNP-SNP Correlations (HAPMAP)
CEU SNP-SNP Correlations (HAPMAP)
CHI SNP-SNP Correlations (HAPMAP)
EAS SNP-SNP Correlations (HAPMAP)
FIN SNP-SNP Correlations (HAPMAP)
INDPAK SNP-SNP Correlations (HAPMAP)
SAS SNP-SNP Correlations (HAPMAP)
UK SNP-SNP Correlations (HAPMAP)
Having downloaded the zip file, you should extract it. On a UNIX operating system, you can do this by typing unzip followed by the file name.
You should now have five files, each starting with HAPMAP.XXX, where XXX is the name of the population. For example, if you downloaded the first set of correlations, you will have the files HAPMAP.AFR.cors.annotations, HAPMAP.AFR.cors.bim, HAPMAP.AFR.cors.bin, HAPMAP.AFR.cors.root and HAPMAP.AFR.cors.windows.
Please note that the file that end with .annotations contains SNP annotations derived from the Baseline LD Model (version 2.2), developed by Marquez-Luna et al.
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Construct the prediction model:
The main argument is --quick-prs <outfile>.
This requires two options
--summary <sumsfile> - to specify the file containing the summary statistics.
--cors <corstem> - to specify the predictor-predictor correlations.
By default, LDAK will use the SNP annotations provided within the correlation files; to instead ignore them, add --use-annotations NO. If the summary statistics file contains A1 frequencies, LDAK will compute the difference between the frequencies in the summary statistics and those in the correlations, then exclude predictors for which the difference is greater than 0.2 (you can stop this by adding --check-frequencies NO).
Most likely, you will be missing some SNPs, in which case you should add --check-sums NO. Note that, by default, LDAK ignores alleles with ambiguous alleles (those with alleles A & T or C & G) to protect against possible strand errors. If you are confident that these are correctly aligned, you can force LDAK to include them by adding --allow-ambiguous YES.
You can use --model <type> to specify which version of QuickPRS to run. By default, LDAK will run BayesR-SS (equivalent to adding --model bayesr), which is our recommended version. The other versions are Lasso-SS (--model lasso), Lasso-SS-Sparse (--model lasso-sparse), Ridge-SS (--model ridge), Bolt-SS (--model bolt), BayesR-SS-Shrink (--model bayesr-shrink) and Elastic-SS (--model elastic). Meanwhile, if you use --model mega, QuickPRS will run Lasso-SS, Ridge-SS, BayesR-SS AND Elastic-SS.
By default, LDAK will use four MCMC chains, each with 250 iterations, the first 50 of which are discarded; you can change these values using --MCMC-chains, --MCMC-iterations and --MCMC-burn. If you wish to compute PRS standard deviations, you should add --PRS-variance YES (LDAK will then save effect sizes every tenth iteration; change this frequency using --MCMC-step).
The estimated effect sizes will be saved in <outfile>.effects. This file will usually have five columns, providing the predictor name, its A1 and A2 alleles, the average number of A1 alleles, then its estimated effect (relative to the A1 allele), and is ready to be used for Calculating Scores (i.e., to predict the phenotypes of new samples). Note that if you used --PRS-variance YES, then <outfile>.effects will contain extra columns corresponding to the MCMC samplings.
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Example:
Here we use the summary statistics file quant.summaries, created in the example for Single-Predictor Analysis. Note that these were created by downloading the Test Datasets, then running the command
./ldak.out --linear quant --bfile human --pheno quant.pheno
Please note that this example is very simple; a more realistic example is provided in Worked Examples.
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
1 - Preparation
The first step is to ensure the summary statistics are in the format required by LDAK, and that the SNP names are rsIDs. By viewing the top of the file quant.summaries, we see that this is already the case
head -n 5 quant.summaries
Predictor A1 A2 Z n A1Freq
rs2260895 A G 0.8750 424 0.4422
rs2821796 A C -1.2416 424 0.4870
rs1148737 G A 0.4353 424 0.4847
rs1296972 T C 0.2084 424 0.4917
Next we download the SNP details file, and use these to work out the overlap between the summary statistics and the HapMap3 SNPs. Note that the following command works because the SNP names are in Column 1 of quant.summaries (if they were instead in, say, Column 2, you should replace "arr[$1]" with "arr[$2]").
wget https://genetics.ghpc.au.dk/doug/Correlations/HAPMAP.SNP.details
awk '(NR==FNR){arr[$1];next}($1 in arr){b++}END{print "HAPMAP: " b, "out of", FNR, b/FNR*100"%"}' quant.summaries HAPMAP.SNP.details
The screen output tells us that we have very low overlap with the HapMap3 SNPs (only 3171 of the 1.1M SNPs). Normally, we should stop and switch to MegaPRS (as explained above, the overlap should ideally be at least 80%). However, because this is only an example, we will continue using the pre-computed SNP-SNP correlations.
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
2 - Download and extract pre-computed SNP-SNP correlations
The test datasets contain European samples, and therefore we should use UK pre-computed SNP-SNP correlations. Therefore, we download the files with prefix HAPMAP.UK
wget https://genetics.ghpc.au.dk/doug/Correlations/HAPMAP.UK.zip
unzip HAPMAP.UK.zip
We should now the following five files:
HAPMAP.UK.cors.annotations
HAPMAP.UK.cors.bim
HAPMAP.UK.cors.bin
HAPMAP.UK.cors.root
HAPMAP.UK.cors.windows
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
3 - Construct the prediction model
Normally, we would estimate effect sizes using the command
./ldak.out --quick-prs quick --summary quant.summaries --cors HAPMAP.UK --check-sums NO
However, this command will fail due to the small number of SNPs (relative to the number of categories in the Baseline LD Model). Therefore, we instead run the command
./ldak.out --quick-prs quick --summary quant.summaries --cors HAPMAP.UK --check-sums NO --use-annotations NO
The estimated effect sizes are saved in the file quick.effects.
