A0660
Title: Cauchy combination test with thresholding under arbitrary dependency structures
Authors: Woncheol Jang - Seoul National University (Korea, South) [presenting]
Abstract: Combining individual p-values to aggregate sparse and weak effects is of substantial interest in large-scale data analysis. Although many p-value combination methods are developed under the i.i.d. assumption, individual p-values or test statistics are often correlated. The Cauchy combination test is a popular method for combining p-values under an arbitrary dependence structure; however, in practice, its Type I error increases as correlation strength increases. A global test that extends the Cauchy combination test is proposed by thresholding and rescaling arbitrarily dependent p-values. By filtering a set of correlated, non-significant p-values, it is demonstrated that the tail probability of the proposed method remains asymptotically equivalent to that of the Cauchy distribution while effectively controlling Type I error under an arbitrary dependence structure. Additionally, it is proven that the power of the proposed test asymptotically achieves the optimal detection boundary under a strong sparsity condition. Extensive simulation results show that the power of the proposed test is robust to different correlation structures and exhibits superior performance in sparse settings. As a case study, the proposed test is applied to a genome-wide association study (GWAS) of inflammatory bowel disease (IBD).
