Robocov

Robocov estimates sparse covariance and inverse covariance matrices for gene-level tissue networks from gene expression data with missing entries using a convex optimization framework.

Key Features:

• Convex Optimization Framework: Employs a convex optimization approach to learn covariance and inverse covariance matrices in the presence of missing data.
• Sparse Estimation: Produces sparse estimates for covariance and inverse covariance matrices to enable inference of conditional dependencies and partial correlations.
• Improved Interpretability: Provides clearer correlation and partial correlation representations for interpretation of correlation and causal structures.
• Robust Missing-Data Handling: Specifically addresses missing entries common in datasets such as the Genotype-Tissue Expression (GTEx) Project when estimating correlation structures.
• Reduced False Positives: Achieves a lower false positive rate compared to traditional correlation estimators and competing approaches such as CorShrink.

Scientific Applications:

• Gene Expression Analysis: Analyzes gene expression across multiple tissues to identify genes with correlated expression informative of systemic genetic activity.
• Pathway Enrichment: Prioritized genes based on Robocov correlations or partial correlations are enriched for pathways related to signaling, heat stress response, immune function, and circadian rhythms.
• Genetic Insights into Blood-Related Traits: Links SNPs associated with prioritized genes to signals relevant for blood-related traits and potential autoimmune disease architectures.
• Systemic Genetic Activity and Disease Architecture: Enables studies of systemic activities across tissues and contributes to understanding genetic and autoimmune disease architectures.

Methodology:

Uses a convex optimization framework to robustly learn sparse covariance and inverse covariance matrices from incomplete gene expression datasets (e.g., GTEx) and reports lower false positive rates versus traditional estimators and CorShrink.