A multivariate statistical test for differential expression analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Statistical tests for differential gene expression often lack power with small, skewed datasets.
• Current methods frequently discretize gene expression data using arbitrary thresholds, potentially losing information.

Purpose of the Study:

• Introduce the Hy-test, a novel statistical method based on multivariate hypergeometric distributions.
• Address limitations in statistical power and arbitrary data discretization in differential expression analysis.

Main Methods:

• Developed the Hy-test using a convolution of multivariate hypergeometric distributions.
• Applied and compared Hy-test to transcriptomic data from breast and kidney cancer.
• Evaluated Hy-test's performance against existing differential expression analysis methods.

Main Results:

• Hy-test demonstrated improved selectivity in identifying differentially expressed genes.
• The method implicitly discretizes expression profiles, enhancing accuracy.
• Successfully retrieved relevant Gene Ontology terms, such as cell cycle deregulation in breast cancer and programmed cell death in kidney cancer.

Conclusions:

• Hy-test offers a robust solution for differential gene expression analysis, particularly for challenging datasets.
• It enhances the discovery of biologically relevant genes and pathways in cancer.
• Hy-test can complement existing methods to uncover hidden biological insights.