DANUBE: Data-driven meta-ANalysis using UnBiased Empirical distributions-applied to biological pathway analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Identifying significantly impacted biological pathways in phenotypes is challenging due to patient heterogeneity and limited statistical power in individual studies.
• Classical meta-analysis methods assume p-values follow a uniform distribution under the null hypothesis, an assumption violated by mainstream pathway analysis methods, leading to biased results.

Purpose of the Study:

• To introduce DANUBE, a novel and unbiased framework for combining statistics from individual studies to improve pathway enrichment analysis.
• To address the limitations of classical meta-analysis in pathway identification by correcting for biased p-value distributions.

Main Methods:

• DANUBE constructs empirical null distributions using control samples to calculate empirical p-values from individual studies.
• Empirical p-values are combined using either a Central Limit Theorem approach or an additive method.
• The performance of DANUBE was assessed using four pathway analysis methods across 16 datasets for Alzheimer's disease and acute myeloid leukemia, comparing it against five meta-analysis approaches and MetaPath.

Main Results:

• DANUBE demonstrates consistent identification of relevant pathways by overcoming the bias present in classical meta-analysis approaches.
• The framework shows improved results compared to common experiment-level statistical tests like Wilcoxon and t-test when applied in a meta-analysis context.
• DANUBE effectively identifies significantly impacted pathways even with limited sample sizes and inherent noise in biological data.

Conclusions:

• DANUBE provides a robust and unbiased solution for pathway enrichment meta-analysis, enhancing the reliability of findings in complex diseases.
• The developed framework significantly improves the accuracy and power of pathway identification compared to existing meta-analysis techniques.
• DANUBE's ability to correct for biased p-value distributions makes it a valuable tool for large-scale biological data integration and interpretation.