GWAS summary statistics and code for "Learning lifetime disease liability reveals and removes genetic confounding in electronic health records"

GWAS on nine common diseases in three different phenotype definitions in UKBiobank, including two psychiatric disorders, generalized anxiety disorder (GAD) and major depressive disorder (MDD), as well as seven non-psychiatric diseases: anaemia (ANEM), chronic obstructive pulmonary disease (COPD), hyper-LDL-cholesterolemia (HLDL), non-alcoholic fatty liver disease (NAFLD), diabetes (DBT), hypertension (HTN), and osteoporosis (OST). Deep phenotype-based: groundtruth.allchr.{disease}.glm.logistic.gz EHR phenotype-based: ICD.allchr.{disease}.glm.logistic.gz predicted disease liability: pred.allchr.{disease}.glm.linear.gz Common_Bias.txt.gz is the common factor of nine disease biases, where each disease bias is derived from GWAS-by-Subtraction to isolate the genetic signal present in the EHR phenotype but independent of the deep phenotype. "src.zip" contains: AutoICD2 includes codes for all deep learning and active learning experiments based on the adapted AutoComplete model. EHRSeq includes codes for all transformer-based experiments. nonwhitebrtishPRS and PRS_Pleotropy include codes for all PRS and PRS Pleiotropy analyses. gsem includes codes for deriving disease-specific biases (using GWAS-by-Subtraction) and the Common Bias model. gsem_bias_correction and External_replication include codes for bias removal (using GWAS-by-Subtraction) analyses.