|
Abstract:
|
Complex phenotypes are the results of coordinated activity of a group of biomolecules. Therefore, the study of biomarker-sets (e.g., SNPs in a gene or genes in a function) are essential for better understanding the biological mechanisms underlying phenotypic variations. Marker-set analysis can be roughly classified into two flavors: (a) “meta”-based approach that evaluates the association of each marker and then integrates summary statistics for an multi-variant association, or (b), a “joint-modeling” that regresses the outcome on all markers simultaneously. The general consensus is that the joint modeling is the “golden standard,” while the meta-approach is at best an approximation. However, joint-modeling demands full access to all data at analytical time, posing logistic issues in contemporary consortia studies where data sharing is restricted and privacy is of concern, while the meta-based approach benefits from separation of data and reports, and the ability to overcome high dimension via divide and conquer. Here, we show equivalences between meta- and joint- approaches, and highlight “hidden treasures” of working with summary statistics.
|
