.png){kind=icon}
Return
FDR Control for High-Dimensional Graphical Models via e$$ e $$-Values
.png){kind=icon}
PRE
AI
DOI：10.1002/sta4.70123.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Graphical models are indispensable for capturing complex dependencies in high-dimensional data across bioinformatics, neuroscience and social sciences. Despite substantial methodological advances in graphical modelling, false discovery rate (FDR) control for graph structure inference remains challenging due to pervasive dependence among test statistics, which violates assumptions of traditional -value-based methods. In this paper, we propose a new FDR control method for high-dimensional graphical models via -values (GEBH). By leveraging -values, the key advantage is that our method can accommodate arbitrary dependence structures among test statistics. Building on GEBH, we further develop GEBH exploiting the partially penalized regression (GEBH-P), which applies to a wider range of graphical applications by relaxing the sparsity constraints in the underlying graph. Theoretical guarantees are established for both methods. The usefulness of our methods is validated via simulations and real data analysis.
Keywords:
e$$ e $$-values
false discovery rate
graphical models
high-dimensionality
partially penalized regression
Journal
S
IF:
0.8
Papers:
59
Citations:
655
Organization
Cited Papers
ShareSaveAdjusting the Benjamini–Hochberg method for controlling the false discovery rate in knockoff-assisted variable selection.png)
Biometrika
OAAI
BiometrikaIF0
OAAIShareSaveShareSave
View more