Gene-set distance analysis (GSDA): a powerful tool for gene-set association analysis.

Area of Science:

• Bioinformatics
• Systems Biology
• Statistical Genetics

Background:

• Identifying gene sets associated with phenotypes or treatments provides biological insights.
• Existing methods often detect simple monotonic gene-gene relationships, missing complex non-monotonic associations.
• Distance correlation methods can detect non-monotone associations but need generalization for categorical and event-time data and methods to identify driving genes.

Purpose of the Study:

• To generalize distance correlations for gene-set association analysis with categorical and censored event-time variables.
• To develop a method for identifying specific genes that drive significant gene-set associations.
• To evaluate the performance of the new method against existing approaches.

Main Methods:

• Gene-set distance analysis (GSDA) was developed by generalizing distance correlations.
• A backward elimination procedure was incorporated to identify key driver genes within significant gene sets.
• The method was validated using simulation studies and a real-world pediatric acute myeloid leukemia (AML) dataset.

Main Results:

• GSDA demonstrated superior performance in identifying complex non-monotone gene-set associations compared to six other methods in simulations.
• GSDA uniquely identified an association between an AML gene set and event-free survival (EFS) in a pediatric AML cohort.
• The method successfully narrowed down the association to 5 key genes and validated this finding in a separate cohort.

Conclusions:

• GSDA is a robust and versatile tool for detecting gene-set associations across various data types (categorical, quantitative, censored event-time).
• The method excels at uncovering complex non-monotonic gene-set associations often overlooked by traditional approaches.
• GSDA facilitates a deeper understanding of gene-set contributions to biological outcomes and is available as an R package.