Improving data interpretability with new differential sample variance gene set tests.

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Current gene set analysis methods primarily detect mean expression changes in omics data.
• Methods for detecting variance changes are underexplored but crucial for understanding biological heterogeneity.
• Existing approaches overlook gene sets with subgroup-specific changes within phenotypes, hindering biological interpretation.

Purpose of the Study:

• To develop and validate multivariate sample-level variance analysis methods for gene set analysis.
• To address the limitations of existing methods in detecting heterogeneous pattern changes.
• To provide novel tools for interpreting omics data with complex biological differences.

Main Methods:

• Generalized univariate statistics (Cramer-Von Mises, Anderson-Darling) into multivariate methods using minimum spanning tree ranking.
• Developed methods to detect differential sample variance and mean.
• Evaluated detection power and Type I error rates via simulations and real-world datasets.

Main Results:

• Applied methods to gene expression datasets from acute lymphoblastic leukemia and colorectal polyps.
• Demonstrated that variance analysis methods identify biologically relevant gene sets missed by mean-based methods.
• Successfully detected hallmark gene sets associated with distinct phenotypes and subtypes.

Conclusions:

• Methods detecting differential sample variance offer significant biological interpretations for heterogeneous data.
• These new approaches are valuable for understanding complex signaling pathways and phenotype differences.
• Software implementation (GSAR package) is available for gene expression and other omics data.