High-sensitivity pattern discovery in large, paired multiomic datasets.

Area of Science:

• Computational biology
• Bioinformatics
• Statistical genetics

Background:

• Biological screens generate vast amounts of data, necessitating robust methods for identifying significant associations.
• Analyzing multiple high-dimensional datasets from the same samples requires approaches with high statistical power and controlled false discovery rates (FDR).

Purpose of the Study:

• To introduce HAllA (Hierarchical All-against-All association testing), a novel framework for structured association discovery between paired high-dimensional datasets.
• To enable the identification of significant linear and non-linear relationships within and between continuous and categorical data.

Main Methods:

• HAllA integrates hierarchical hypothesis testing with FDR correction for efficient association discovery.
• The framework was optimized and evaluated using synthetic datasets with known structures.
• Performance was compared against all-against-all and other block-testing methods using various similarity measures.

Main Results:

• HAllA demonstrated superior performance compared to existing methods on synthetic datasets.
• Application to real-world multi-omics data revealed novel associations between gene expression and immune activity.
• New links were identified between the microbiome and host transcriptome, and between metabolomics and human health phenotypes.

Conclusions:

• HAllA provides a powerful and flexible framework for uncovering complex biological associations in multi-omics data.
• The method offers improved statistical power and FDR control for structured association discovery.
• HAllA facilitates the exploration of relationships across diverse biological datasets, advancing multi-omics research.