Extending gene set variation analysis with a reference dataset to stabilize scores.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene set variation analysis (GSVA) summarizes gene sets into pathway activity scores.
• Current GSVA relies on input data for gene distribution estimation, leading to bias in imbalanced datasets.
• This limitation hinders accurate pathway activity interpretation in studies with non-representative sample distributions.

Purpose of the Study:

• To introduce reference stabilizing GSVA (rsGSVA), a novel method to overcome GSVA's reliance on input data distributions.
• To enhance the stability and reproducibility of gene set enrichment scores.
• To provide a more robust tool for biological pathway analysis.

Main Methods:

• rsGSVA utilizes external reference datasets to estimate gene distributions for calculating pathway scores.
• This approach decouples score calculation from the specific characteristics of the input sample set.
• The method was evaluated for its performance and stability compared to existing techniques.

Main Results:

• rsGSVA demonstrates comparable analytical power to established methods like GSVA, singscore, and ssGSEA in ideal conditions.
• rsGSVA scores remain stable even when analyzing subsets of the input data.
• Application to irritable bowel disease data revealed rsGSVA's advantage in interpreting inflammation-related pathway activity.

Conclusions:

• rsGSVA enhances GSVA by incorporating a reference dataset for more reliable gene distribution estimation.
• The method ensures enrichment scores are independent of the input data's distribution, improving stability.
• rsGSVA offers enhanced reproducibility and interpretability for pathway analysis, particularly with imbalanced or evolving datasets.