This repository contains the code and resources of the following paper:
PRS-Net: Interpretable polygenic risk scores via geometric learning
PRS-Net is an interpretable genomic deep learning-based approach designed to effectively model the nonlinearity of the biological system and deliver precise and interpretable polygenic risk scores.
Setup the required environment using environment_data.yml with Anaconda. While in the project directory run:
conda env create -f environment_data.yml
Activate the environment
conda activate PRS-Net_data
Step 1: Arrange GWAS summary statistic data
Ensure the columns in your GWAS summary statistic data are ordered as follows, separated by tabs (\t): CHR, BP, SNP, A1, A2, N, SE, P, OR, INFO, MAF, BETA, where A1 stands for the effect allele of the SNP and A2 stands for the non-effect allele of the SNP.
Step 2: Run preprocessing script
Execute the 1_data_preprocess.sh script with the following command:
./1_data_preprocess.sh your_base_data_path your_target_data_path your_phenotype_data_path your_ld_data_path your_output_path
-
your_base_data_path: the path of the gwas summarty stastistics file derived from Step 1
-
your_target_data_path: the path (without postfix) of the target data in PLINK format (.bim, .fam, .bed)
-
your_phenotype_data_path: the path of the phenotype data separated by blanks (Sample_ID Sample_ID Phenotype_label)
-
your_ld_data_path: the path (without postfix) of the data used for LD calculating in PLINK format
-
your_output_path: the path of the output files
Step 3: Run PRS-Net data generation script
Execute the gwas_data_preprocess.sh script with the following command:
./2_data_generation.sh your_preprocessed_data_path your_phenotype_data_path your_output_path
-
your_preprocessed_data_path: the path of the preprocessed files derived from Step 2
-
your_phenotype_data_path: the path of the phenotype data separated by tabs (ID \t ID \t Label)
-
your_output_path: the path of the output files
Setup the required environment using environment.yml with Anaconda. While in the project directory run:
conda env create -f environment.yml
Activate the environment
conda activate PRS-Net
We upload an example dataset at https://cloud.tsinghua.edu.cn/f/30a1c8df0c024fefa3c1/?dl=1.
Unzip the file and put it in example_dataset/ and excecute the train.py with the following command:
python train.py --data_path ../example_dataset/ --dataset ad_eur
PRS-Net is licensed under the Apache License, Version 2.0: http://www.apache.org/licenses/LICENSE-2.0.