MIXPRS

MIXPRS is a data fission-based multi-population PRS integration framework designed to effectively combine PRS derived from multiple populations and methods. This framework is described in detail in:

Xu, L., Dong, Y., Zeng, X., Bian, Z., Zhou, G., Guan, L., & Zhao, H. (2025). Almost Free Enhancement of Multi-Population PRS: From Data-Fission to Pseudo-GWAS Subsampling. bioRxiv, 2025-06.

The MIXPRS pipeline requires only GWAS summary statistics and LD reference panels, and involves three main steps (Figure 1):

Step1: GWAS subsampling in MIXPRS
Generate independent subsampled training and tuning GWAS datasets.
Step2: Estimation of MIXPRS combination weights for different PRS methods
Estimate combining weights using two component methods, JointPRS-auto and SDPRX, each requiring only GWAS summary statistics and LD reference panels.
Step3: Derivation of MIXPRS
Calculate the integrated MIXPRS using weights estimated in Step 2.

Detailed implementations of these component methods are available in their respective repositories:

For full details of codes, see the MIXPRS_analysis repository.

Getting Started

In this section, we provide detailed, step-by-step instructions for implementing MIXPRS. Please replace all placeholders with the appropriate paths and filenames specific to your computing environment.

1. MIXPRS Installation

For the first time, you need to use the following code to install MIXPRS:

git clone https://github.com/LeqiXu/MIXPRS.git
cd MIXPRS
conda env create -f environment.yml
conda activate MIXPRS

After this, you only need to use

conda activate MIXPRS

2. LD Reference Panel Download

We use reference panels from PRS-CSx and you can follow their instructions to download them. It is strongly recommended to create two subfolders within your reference directory:

1KG: This subfolder should contain LD reference panels constructed using the 1000 Genomes Project phase 3 samples and the corresponding SNP information file. Specifically, include:
- snpinfo_mult_1kg_hm3
- Population-specific directories (ldblk_1kg_afr, ldblk_1kg_amr, ldblk_1kg_eas, ldblk_1kg_eur, ldblk_1kg_sas), each containing:
  - ldblk_1kg_chr1.hdf5 through ldblk_1kg_chr22.hdf5
  - snpinfo_1kg_hm3
UKBB: This subfolder should contain LD reference panels constructed using the UK Biobank data and the corresponding SNP information file. Specifically, include:
- snpinfo_mult_ukbb_hm3
- Population-specific directories (ldblk_ukbb_afr, ldblk_ukbb_amr, ldblk_ukbb_eas, ldblk_ukbb_eur, ldblk_ukbb_sas), each containing:
  - ldblk_ukbb_chr1.hdf5 through ldblk_ukbb_chr22.hdf5
  - snpinfo_ukbb_hm3

Then place the downloaded LD reference panels and the SNP information file into their corresponding subfolders.

3. Pruned SNP List Preparation

In the MIXPRS repository, we provide precomputed pruned SNP lists for five populations (EUR, EAS, AFR, SAS, and AMR), derived from genotype data of the Phase-3 1000 Genomes Project. These SNP lists were generated using PLINK2 with parameters: window size = 250, step size = 5, and correlation threshold $r^2 = 0.5$.

To replicate our results or generate customized pruned SNP lists, use the following PLINK2 commands:

module load PLINK/2

i=1
pval=1
r2=0.5
wc=250

for pop in EUR EAS AFR SAS AMR; do

plink2 --bfile ./1000g_phase3_data/geno_data/${pop} \
       --indep-pairwise ${wc} 5 ${r2} \
       --out ./snplist/${pop}_prune_pval${pval}_r2${r2}_wc${wc}_${i}
done

You can adjust these parameters or apply the commands to other genotype datasets according to your analytical needs.

4. Summary Statistics Preparation

We require GWAS summary statistics provided as input to MIXPRS to follow the format below (including the header line):

SNP         A1  A2  BETA        SE          Z                P      N
rs3934834   T   C   0.00492590  0.00761858  0.646564057869   0.518  100061
rs3766192   C   T   0.00144597  0.00580197  0.249220523374   0.803  100953
rs9442372   A   G   0.00261076  0.00586576  0.445084694907   0.656  98059
...

Here

SNP: SNP rsID.
A1: Effect allele.
A2: Alternative allele.
BETA: Effect size of allele A1, which is only used to determine the direction of an association.
SE: Standard error of the effect size.
Z: Z-score (BETA divided by SE).
P: P-value of the effect size, which is used to calculate the standardized effect size.
N: GWAS sample size.

5. MIXPRS Implementation

Step 0: MIXPRS data preparation

Before running MIXPRS, ensure these datasets are ready (see previous sections):

LD Reference Panel (Section 2).
Pruned SNP List (Section 3).
GWAS Summary Statistics (Section 4).

Example assumptions:

1KG LD panel used.
Precomputed pruned SNP lists from MIXPRS repository (snplist).
Formatted GWAS summary statistics ${trait}_${pop}_MIXPRS_sumstat.txt. We provide an example summary statistics dataset for EAS HDL (500 SNPs on chromosome 1) within the cloned MIXPRS repository (example_data folder), obtained from GLGC.

Step1: GWAS subsampling in MIXPRS

This step partitions a single original GWAS dataset into independent subsampled training and tuning GWAS datasets using data fission principles for the target population (pop).

The default parameters for subsampling are:

Requires a pruned SNP list (--prune_snplist) provided in the MIXPRS repository (snplist folder).
indep_approx=TRUE: uses an identity covariance matrix for the pruned SNPs instead of LD reference panels (the --ref_dir flag is still required).
train_tune_ratio=3: divides the original GWAS dataset into subsampled training GWAS (3/4 of samples) and subsampled tuning GWAS (1/4 of samples).
repeat=4: generates four independent subsampled training-tuning GWAS pairs for robust evaluation.

Run the following command to subsample training and tuning GWAS:

conda activate MIXPRS

python ${MIXPRS_path}/MIX_subsample2.py \
  --ref_dir=${ref_data_path}/1KG \
  --sst_file=${summary_stat_path}/${trait}_${pop}_MIXPRS_sumstat.txt \
  --pop=${pop} \
  --prune_snplist=${MIXPRS_path}/snplist/${pop}_prune_pval1_r20.5_wc250_1.snplist \
  --indep_approx=TRUE \
  --train_tune_ratio=3 \
  --repeat=4 \
  --out_dir=${output_path}/subsample/clean \
  --out_name=${trait}_prune_snplist_1

This command generates four pairs of independent subsampled training and tuning GWAS summary statistics files, named as follows (repeat=1,2,3,4):

Subsampled training GWAS: ${output_path}/subsample/clean/${trait}_prune_snplist_1_${pop}_train_GWAS_approxTRUE_ratio3.00_repeat${repeat}.txt
Subsampled tuning GWAS: ${output_path}/subsample/clean/${trait}_prune_snplist_1_${pop}_tune_GWAS_approxTRUE_ratio3.00_repeat${repeat}.txt

Example format of these generated subsampled GWAS files:

Subsampled training GWAS example:

SNP         CHR     BP      A1  A2  A1_Frq      BETA           SE           Z         P         N
rs3131969   1       754182  A   G   2.669e-01  -3.482222e-03  7.831456e-03 -4.446456e-01  6.565759e-01  74586
rs1048488   1       760912  C   T   1.161e-01  -3.537603e-03  7.797450e-03 -4.536872e-01  6.500540e-01  85875
rs1806509   1       853954  C   A   4.137e-01   4.359470e-03  7.256207e-03  6.007918e-01  5.479786e-01  60434
...

Subsampled tuning GWAS example:

SNP         CHR     BP      A1  A2  A1_Frq      BETA           SE           Z         P         N
rs3131969   1       754182  A   G   2.669e-01   2.987062e-02  1.356448e-02  2.202121e+00  2.765678e-02  24862
rs1048488   1       760912  C   T   1.161e-01   1.037583e-02  1.350558e-02  7.682627e-01  4.423312e-01  28625
rs1806509   1       853954  C   A   4.137e-01   1.163264e-03  1.256812e-02  9.255673e-02  9.262557e-01  20144
...

Step2: Estimation of MIXPRS combination weights for different PRS methods

This step includes two substeps:

Step 2.1: Obtain LD-pruned PRS beta files

Use the subsampled training GWAS for the target population obtained from Step 1.
For GWAS from other populations, apply the LD-pruned SNP list from the target population (instead of their original populations) to filter and obtain LD-pruned GWAS summary statistics from their original GWAS summary statistics.
Format these aligned, pruned GWAS summary statistics from all populations for each method and implement each method according to their respective repositories:
- JointPRS-auto
- SDPRX

Repeat this procedure for each of the four subsampled training GWAS sets generated in Step 1 (repeat=1,2,3,4). After completing this step, you should obtain beta files similar to the following for each repeat (repeat=1,2,3,4):

${JointPRS_EUR_beta_file}, ${JointPRS_EAS_beta_file}, ${JointPRS_AFR_beta_file}, ${JointPRS_SAS_beta_file}, ${JointPRS_AMR_beta_file}
${SDPRX_EUR_beta_file}, ${SDPRX_EAS_beta_file}, ${SDPRX_AFR_beta_file}, ${SDPRX_SAS_beta_file}, ${SDPRX_AMR_beta_file}

Note: The number of population-specific beta files obtained depends on the availability of GWAS summary statistics for each population.

Step 2.2: Obtain PRS combining weights

Use the subsampled tuning GWAS and format them according to the required MIXPRS summary statistics format detailed in Section 4. We assume the formatted subsampled tuning GWAS file is named as follows: ${output_path}/subsample/MIXPRS/${trait}_prune_snplist_1_${pop}_tune_MIXPRS_approxTRUE_ratio3.00_repeat${repeat}.txt
Use the LD-pruned PRS (beta files) obtained from Step 2.1.
Default parameters for this step are:
- indep_approx=TRUE: uses an identity covariance matrix for the pruned SNPs instead of LD reference panels (the --ref_dir flag is still required).
- selection_criterion=NNLS: employs the Lawson–Hanson Non-Negative Least Squares (NNLS) algorithm to estimate non-negative PRS combining weights.

Run the following command to calculate the optimal PRS combining weights:

conda activate MIXPRS

python ${MIXPRS_path}/MIX_linear_weight.py \
  --ref_dir=${ref_data_path}/1KG \
  --sst_file=${output_path}/subsample/MIXPRS/${trait}_prune_snplist_1_${pop}_tune_MIXPRS_approxTRUE_ratio3.00_repeat${repeat}.txt \
  --pop=${pop} \
  --prs_beta_file=${JointPRS_EUR_beta_file},${JointPRS_EAS_beta_file},${JointPRS_AFR_beta_file},${JointPRS_SAS_beta_file},${JointPRS_AMR_beta_file},${SDPRX_EUR_beta_file},${SDPRX_EAS_beta_file},${SDPRX_AFR_beta_file},${SDPRX_SAS_beta_file},${SDPRX_AMR_beta_file} \
  --indep_approx=TRUE \
  --selection_criterion=NNLS \
  --out_dir=${output_path}/Final_weight/no_val \
  --out_name=${trait}_prune_snplist_1_JointPRS_SDPRX_EUR_EAS_AFR_SAS_AMR_repeat${repeat}

Repeat this procedure for each of the four subsampled tuning GWAS sets generated in Step 1 (repeat=1,2,3,4). After completing this step, you should obtain four PRS combining weights files, named as follows for each repeat (repeat=1,2,3,4):

PRS combining weights: ${output_path}/Final_weight/no_val/${trait}_prune_snplist_1_JointPRS_SDPRX_EUR_EAS_AFR_SAS_AMR_repeat${repeat}_${pop}_non_negative_linear_weights_approxTRUE.txt

Note: For each repeat (repeat=1,2,3,4), ensure you use the corresponding subsampled tuning GWAS dataset from Step 1 along with its matching LD-pruned PRS beta file obtained from Step 2.1.

Step3: Derivation of MIXPRS

This step includes two substeps:

Step 3.1: Obtain full SNPs PRS beta files

Use the original GWAS summary statistics from all populations.
Format these original GWAS summary statistics according to the requirements of each method and implement each method following their respective repositories:
- JointPRS-auto
- SDPRX

After completing this step, you should obtain full SNPs beta files for each method and population, such as:

${JointPRS_EUR_beta_file}, ${JointPRS_EAS_beta_file}, ${JointPRS_AFR_beta_file}, ${JointPRS_SAS_beta_file}, ${JointPRS_AMR_beta_file}
${SDPRX_EUR_beta_file}, ${SDPRX_EAS_beta_file}, ${SDPRX_AFR_beta_file}, ${SDPRX_SAS_beta_file}, ${SDPRX_AMR_beta_file}

Note: The number of population-specific beta files depends on the availability of GWAS summary statistics for each population.

Step 3.2: Obtain MIXPRS

Use the original GWAS summary statistics for the target population formatted according to Section 4: Summary Statistics Preparation. This should match the file used in Step 1:
```
${summary_stat_path}/${trait}_${pop}_MIXPRS_sumstat.txt
```

Use the calculated PRS combining weights obtained in Step 2.2, named as follows:

weight_file1="${output_path}/Final_weight/no_val/${trait}_prune_snplist_1_JointPRS_SDPRX_EUR_EAS_AFR_SAS_AMR_repeat1_${pop}_non_negative_linear_weights_approxTRUE.txt"
weight_file2="${output_path}/Final_weight/no_val/${trait}_prune_snplist_1_JointPRS_SDPRX_EUR_EAS_AFR_SAS_AMR_repeat2_${pop}_non_negative_linear_weights_approxTRUE.txt"
weight_file3="${output_path}/Final_weight/no_val/${trait}_prune_snplist_1_JointPRS_SDPRX_EUR_EAS_AFR_SAS_AMR_repeat3_${pop}_non_negative_linear_weights_approxTRUE.txt"
weight_file4="${output_path}/Final_weight/no_val/${trait}_prune_snplist_1_JointPRS_SDPRX_EUR_EAS_AFR_SAS_AMR_repeat4_${pop}_non_negative_linear_weights_approxTRUE.txt"

Use the full SNPs PRS beta files obtained from Step 3.1.

Run the following command to obtain the final MIXPRS:

conda activate MIXPRS

python ${MIXPRS_path}/MIX_final_combine.py \
  --ref_dir=${ref_data_path}/1KG \
  --sst_file=${summary_stat_path}/${trait}_${pop}_MIXPRS_sumstat.txt \
  --pop=${pop} \
  --prs_beta_file=${JointPRS_EUR_beta_file},${JointPRS_EAS_beta_file},${JointPRS_AFR_beta_file},${JointPRS_SAS_beta_file},${JointPRS_AMR_beta_file},${SDPRX_EUR_beta_file},${SDPRX_EAS_beta_file},${SDPRX_AFR_beta_file},${SDPRX_SAS_beta_file},${SDPRX_AMR_beta_file} \
  --weight_file=${weight_file1},${weight_file2},${weight_file3},${weight_file4} \
  --out_dir=${output_path}/MIXPRS \
  --out_name=${trait}

This command produces two results files in the specified output directory (${output_path}/MIXPRS):

Final MIXPRS beta:

${output_path}/MIXPRS/${trait}_${pop}_MIXPRS.txt

Weighted PRS beta for each method and population:

${output_path}/MIXPRS/${trait}_${pop}_MIXPRS_separate.txt

Note: Ensure the order and naming of --prs_beta_file are consistent between Step 2.2 and Step 3.2.

Acknowledgment

Part of the code is adapted from PRS-CSx. We thank Dr. Tian Ge for sharing his code and LD reference panels.

Support

Please direct any problems or questions to Leqi Xu ([email protected]).

Name		Name	Last commit message	Last commit date
Latest commit History 58 Commits
example_data		example_data
snplist		snplist
MIX_final_combine.py		MIX_final_combine.py
MIX_linear_weight.py		MIX_linear_weight.py
MIX_subsample2.py		MIX_subsample2.py
README.md		README.md
environment.yml		environment.yml
gwas_subsample.py		gwas_subsample.py
parse_genet.py		parse_genet.py
prs_combine.py		prs_combine.py
prs_weight.py		prs_weight.py

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Repository files navigation

MIXPRS

Getting Started

1. MIXPRS Installation

2. LD Reference Panel Download

3. Pruned SNP List Preparation

4. Summary Statistics Preparation

5. MIXPRS Implementation

Step 0: MIXPRS data preparation

Step1: GWAS subsampling in MIXPRS

Step2: Estimation of MIXPRS combination weights for different PRS methods

Step 2.1: Obtain LD-pruned PRS beta files

Step 2.2: Obtain PRS combining weights

Step3: Derivation of MIXPRS

Step 3.1: Obtain full SNPs PRS beta files

Step 3.2: Obtain MIXPRS

Acknowledgment

Support

About

Uh oh!

Releases

Packages

Languages

YCSGP/MIXPRS

Folders and files

Latest commit

History

Repository files navigation

MIXPRS

Getting Started

1. MIXPRS Installation

2. LD Reference Panel Download

3. Pruned SNP List Preparation

4. Summary Statistics Preparation

5. MIXPRS Implementation

Step 0: MIXPRS data preparation

Step1: GWAS subsampling in MIXPRS

Step2: Estimation of MIXPRS combination weights for different PRS methods

Step 2.1: Obtain LD-pruned PRS beta files

Step 2.2: Obtain PRS combining weights

Step3: Derivation of MIXPRS

Step 3.1: Obtain full SNPs PRS beta files

Step 3.2: Obtain MIXPRS

Acknowledgment

Support

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Languages

Packages