High-dimensional variable selection for ordinal outcomes with error control.

Area of Science:

• Genomics
• Biostatistics
• Bioinformatics

Background:

• High-throughput genomic studies often involve numerous covariates and ordinal responses, necessitating robust variable selection methods.
• Controlling the false discovery rate (FDR) while maintaining statistical power is a critical challenge in identifying relevant variables.

Purpose of the Study:

• To develop and evaluate novel variable importance measures for ordinal responses within existing selection frameworks.
• To compare the performance of model-X knockoffs and reference distribution variable selection (RDVS) in terms of FDR and power.

Main Methods:

• Constructed importance measures for ordinal responses using penalized regression and machine learning techniques.
• Integrated these measures into the model-X knockoffs and RDVS variable selection frameworks.
• Evaluated performance using simulated data and applied to high-throughput methylation data.

Main Results:

• Demonstrated the applicability of penalized regression and machine learning for creating importance measures suitable for ordinal responses.
• Compared the FDR and power of the two frameworks across different scenarios.
• Identified features associated with liver cancer progression using methylation data.

Conclusions:

• The study provides a comparative analysis of two variable selection frameworks for high-throughput genomic data with ordinal outcomes.
• The developed methods offer valuable tools for identifying significant variables in complex biological datasets.
• The findings have implications for understanding disease progression, such as from normal liver tissue to hepatocellular carcinoma.