PLSKO: a robust knockoff generator to control false discovery rate in omics variable selection.

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• The knockoff framework provides False Discovery Rate (FDR) control for high-throughput omics data analysis without p-values.
• Existing knockoff generators often fail in real-world data due to restrictive assumptions or approximations, leading to inflated FDR.

Purpose of the Study:

• Introduce Partial Least Squares Knockoff (PLSKO), an efficient and assumption-free knockoff generator.
• Evaluate PLSKO's performance across diverse omics platforms and complex data settings.
• Improve power and biological relevance in multi-omics studies.

Main Methods:

• Developed Partial Least Squares Knockoff (PLSKO), an assumption-free knockoff generator.
• Conducted extensive simulations and semi-simulation studies using RNA-seq, proteomics, metabolomics, and microbiome data.
• Combined PLSKO with Aggregation Knockoff for enhanced power in multi-omics case studies.

Main Results:

• PLSKO demonstrated robust FDR control and high power in complex, non-linear settings across various omics platforms.
• Semi-simulation studies confirmed the validity of knockoff variables generated by PLSKO.
• The combined PLSKO and Aggregation Knockoff approach successfully identified biologically meaningful features in pre-eclampsia multi-omics data.

Conclusions:

• PLSKO is a powerful and reliable tool for FDR-controlled variable selection in diverse omics data.
• The assumption-free nature of PLSKO enhances its applicability to real-world biological data.
• PLSKO facilitates the discovery of significant biological features in complex multi-omics analyses.