All Times ET
Keywords: meta-analysis, Fisher’s p-value combination, empirical null distribution, Weighted Z statistic, simultaneous hypothesis testing
Recent high-throughput technology developments have enabled simultaneous analysis of numerous genes in a single experiment. For meta-analysis of such large-scale experiments, development of novel approaches is crucial due to the fact that the sample sizes of individual experiments are generally small compared to the number of genes being tested. In such large-scale multiple testing frameworks, adhering to the regular statistical assumptions regarding the null distributions of test statistics can potentially lead to incorrect significance testing results and biased inference. This includes a possibility of ending up with gross false discoveries of significant genes. This problem gets worse when one combines large-scale multiple testing p-values from different independent genomic experiments in presence of some hidden confounders. I will show how empirical adjustments of the individual test statistics and p-values from different independent experiments outperform the standard meta-analysis approaches of significance testing, especially in potential presence of hidden confounders, by accurately identifying the truly significant genes in simulation studies and real genomic datasets.