Deciphering New Pathogenic Loci Based On Summary Statistics Of Genome-wide Association Analysis

Genome-wide association analysis has been widely used to study complex diseases.With the development of sequencing technology,the scale of genome-wide association analysis is becoming larger and larger,and more unknown pathogenic loci of complex diseases have been discovered.However,due to the high restrictions of genetic resources and ethical security,only summary statistics data have been released for genome-wide association analysis.Therefore,how to make better use of the summary statistics of genome-wide association analysis is an important issue in the post-genome-wide association analysis era.This dissertation aims to explore more pathogenic loci by performing fine mapping of those identified loci through statistical analysis,supervised learning and graph embedding model based on the summary statistics of genome-wide association analysis.In addition,we explored the causal effect relationship between different traits with Mendelian randomization analysis.Firstly,we performed a multi-trait combination analysis of three psychiatric disorders with high genetic correlation and pleiotropy—schizophrenia,major depression,and bipolar disorder,and identified some new pathogenic loci.We then focused on the pleiotropic effects of FURIN mapped by rs8039305 in three psychiatric disorders,which were validated with the data from GEO database.We also detected the causal effect of bipolar disorder to schizophrenia using multiple methods of Mendelian randomization analysis.Secondly,we constructed a supervised learning model by using the XGBoost algorithm,to integrate causal e QTL data and >5000 features,and generated a new e QTL priority score by comparing the performance of different models and defining the best parameters,which can be used as a prior probability to enhance the fine mapping.We conducted the statistical fine mapping and validated the results in the brain substantia nigra.Finally,we used three different graph embedding methods,including node2 vec,SNP information-based position encoding and metapath2 vec to integrate genome-wide association analysis summary statistics,LD scores,PPI and disease interactions,and predict new pathogenic loci for target diseases by calculating cosine similarity between vectors.Those studies provide new insights for how to make better use of the GWAS summary statistics in genome-wide association analysis.