A Cyclic Permutation Approach to Removing Spatial Dependency between Clustered Gene Ontology Terms

TL;DR

This paper introduces SAGO, a new method to improve gene set analysis by accounting for spatial gene proximity, reducing false enrichments in large genomic regions.

Contribution

SAGO uses cyclic permutation to distinguish spatial gene dependencies from true biological relationships in large genomic loci.

Findings

SAGO reduces misleading enrichments caused by spatial proximity in large genomic domains.

Application to prostate cancer samples with CNA domains removed most false GO term enrichments.

SAGO improves accuracy in identifying biologically relevant gene sets in complex genomic regions.

Abstract

In the intricate field of genomic research, researchers frequently look for the enrichment of genes with a common function. Traditionally, genes are analyzed as if they function independently. However, this assumption may not hold true in large genomic regions, where genes with similar functions exist in close proximity and may influence each other. Our research introduces an advanced method to discern whether the observed patterns in gene groups are due to their spatial closeness, or stem from other biological factors. This approach is particularly crucial in studying large genomic loci, where conventional methods might overlook the nuanced interplay of functionally similar genes. By implementing our technique, we significantly enhance the precision of genomic analyses, particularly in these extensive areas. This advancement is vital as it deepens our understanding of gene interactions within large genomic regions.

Traditional gene set enrichment analysis falters when applied to large genomic domains, where neighboring genes often share functions. This spatial dependency creates misleading enrichments, mistaking mere physical proximity for genuine biological connections. Here we present Spatial Adjusted Gene Ontology (SAGO), a novel cyclic permutation-based approach, to tackle this challenge. SAGO separates enrichments due to spatial proximity from genuine biological links by incorporating the genes’ spatial arrangement into the analysis. We applied SAGO to various datasets in which the identified genomic intervals are large, including replication timing domains, large H3K9me3 and H3K27me3 domains, HiC compartments and lamina-associated domains (LADs). Intriguingly, applying SAGO to prostate cancer samples with large copy number alteration (CNA) domains eliminated most of the enriched GO terms, thus helping to accurately identify biologically relevant gene sets linked to oncogenic processes, free from spatial bias.

Figures