GWAS Summary Statistics for "Improving polygenic risk prediction performance through integrating electronic health records by phenotype embedding"

These genome-wide association summary statistics were generated as part of the study “Improving polygenic risk prediction performance through integrating electronic health records by phenotype embedding.” The analyses leverage electronic health record (EHR)-derived phenotype embeddings to enhance polygenic risk prediction using only GWAS summary data. Five embedding approaches were evaluated: Word2Vec, Word2Vec_PCA, Word2Vec_ICA, GPT_PCA, and GPT_ICA, to capture latent clinical structures within EHR data. Genome-wide association analyses were conducted in the UK Biobank European training population (N = 207,734), focusing on HapMap3 SNPs to ensure well-imputed, high-quality variants suitable for cross-cohort analyses. The resulting summary statistics (SNP, CHR, POS, A1, A2, N, MAF, BETA, SE, Z, P) enable replication and further meta-analyses of embedding-based GWAS results. This dataset supports the development of the EEPRS and MTAG_EEPRS frameworks described in the accompanying manuscript and preprint (doi:https://doi.org/10.1101/2025.08.05.668705).本全基因组关联分析汇总统计量源自一项题为"通过表型嵌入整合电子健康记录以提升多基因风险预测性能"的研究。 本分析仅依托全基因组关联分析（Genome-Wide Association Study, GWAS）汇总数据，借助电子健康记录（EHR）衍生的表型嵌入技术优化多基因风险预测性能。本次研究共评估了五种嵌入方法：Word2Vec、Word2Vec_PCA、Word2Vec_ICA、GPT_PCA及GPT_ICA，以捕捉电子健康记录数据中潜藏的临床结构特征。 全基因组关联分析在英国生物银行欧洲裔训练队列（N=207734）中开展，分析聚焦于HapMap3 单核苷酸多态性（Single Nucleotide Polymorphism, SNPs），以确保获取适配跨队列分析的高质量、填充完善的遗传变异位点。 本次生成的汇总统计量包含（SNP、CHR、POS、A1、A2、N、MAF、BETA、SE、Z、P）字段，可用于基于嵌入技术的全基因组关联分析结果的重复验证与后续元分析。本数据集可支撑伴随发表手稿及预印本（DOI: https://doi.org/10.1101/2025.08.05.668705）中所述的EEPRS与MTAG_EEPRS框架的开发工作。