Statistical significance of variables driving systematic variation in high-dimensional data

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Principal Component Analysis (PCA) captures systematic variation in large-scale genomic data.
• Identifying genomic variables driving this variation is challenging due to overfitting and inflated statistical significance.
• Existing methods struggle to accurately link principal components (PCs) to specific genomic drivers.

Purpose of the Study:

• To introduce a novel statistical method, the jackstraw, for accurately identifying genomic variables associated with latent variables captured by PCA.
• To simplify complex significance testing in genomics and provide a bridge between systematic variation analysis and gene-level significance.

Main Methods:

• The jackstraw method is proposed to assess statistical significance of genomic variables associated with PCs.
• The method's accuracy is validated through simulations across various scenarios.
• An R software package, jackstraw, is available on CRAN for implementation.

Main Results:

• The jackstraw method accurately identifies genomic variables statistically significantly associated with PCs.
• Simulations confirm the method's accurate statistical significance measures.
• Application to yeast cell-cycle data identified cell-cycle regulated genes accurately.
• Analysis of post-trauma patient data revealed greater enrichment of inflammatory gene sets compared to original analyses.

Conclusions:

• The jackstraw method provides an accurate and simplified approach for identifying genomic drivers of systematic variation.
• It enables robust gene-level significance testing in large-scale genomic studies.
• The method enhances biological discovery by linking molecular data to phenotypes, as demonstrated in cell-cycle and trauma studies.