Super-delta

Super-delta performs robust multivariate normalization and differential expression analysis to improve identification of differentially expressed genes (DEGs) while reducing bias and type I error introduced during normalization.

Key Features:

• Multivariate Normalization: Employs a multivariate extension of global normalization that reduces sample variance while minimizing bias from DEGs and outliers.
• Modified t-Test: Uses a modified t-test derived from asymptotic theory for hypothesis testing to enhance statistical power and control type I error rates.
• Bias Minimization: Implements procedures specifically designed to minimize bias introduced by DEGs during normalization.
• Performance Comparison: Demonstrated superior statistical power and stricter type I error control in simulation studies versus global, median-IQR, quantile, and cyclic loess normalization, approximating an oracle test on noise-free datasets.
• Application to Clinical Data: Applied to gene expression datasets from breast cancer patients undergoing neoadjuvant chemotherapy, identifying more DEGs and stronger associations with breast cancer pathways.
• Pathway Analysis: Downstream gene set enrichment analysis showed largely consistent pathway selection across methods but highlighted differences more relevant to breast cancer biology with Super-delta.

Scientific Applications:

• Differential Expression Analysis: Precise identification of DEGs in gene expression studies.
• Pathway and Gene Set Analysis: Improved downstream gene set enrichment analysis and pathway association in transcriptomic studies.
• Clinical Transcriptomics: Analysis of clinical gene expression datasets such as breast cancer neoadjuvant chemotherapy cohorts.
• Between-Group Comparison Problems: Extension to broader between-group comparison analyses and potential application to other data types.

Methodology:

Combines a multivariate extension of global normalization, procedures to minimize DEG- and outlier-induced bias, a modified t-test derived from asymptotic theory, simulation-based comparisons against global, median-IQR, quantile, and cyclic loess normalization, and downstream gene set enrichment analysis on real datasets.