Grace-AKO

Grace-AKO identifies genes associated with clinical outcomes in high-dimensional genomic data while controlling the finite-sample false discovery rate (FDR).

Key Features:

• Graph-Constrained Estimation (Grace): Implements the Grace method to incorporate gene network structures and biological pathway information into variable selection.
• Aggregation of Multiple Knockoffs (AKO): Integrates aggregation of multiple knockoff procedures to improve robustness of finite-sample FDR control.
• Network-Constrained Penalty: Applies a network-constrained penalty so the selection process respects underlying biological networks.
• Improved Model Stability: Demonstrates enhanced stability and tighter finite-sample FDR control relative to the original Grace model in simulation studies.

Scientific Applications:

• Pathway-informed genomic studies: Refines gene selection by incorporating prior biological knowledge such as pathways from the Kyoto Encyclopedia of Genes and Genomes (KEGG).
• Prostate cancer biomarker discovery: Applied to The Cancer Genome Atlas prostate cancer data with prostate-specific antigen (PSA) pathways, identifying 47 candidate genes of which over 75% were validated.

Methodology:

Combines graph-constrained estimation (Grace) with aggregation of multiple knockoffs (AKO) and applies a network-constrained penalty to control finite-sample FDR.