PathwayDenester: a novel method for improving pathway enrichment analyses by taking overlap structures into account

Pathway enrichment analysis is a powerful approach for measuring the overrepresentation of genes of interest in any annotated gene list. It is commonly used to describe a list of selected genes in easily interpretable terms. In current pathway enrichment methods, each pathway is tested independently, without taking into consideration the hierarchical term structure where the largest terms contain several smaller terms that partially intersect each other and in succession contain even smaller terms. Because of that, some highly significant pathways will inevitably share all, or some, genes with other pathways that would be also considered enriched just based on intersected genes. This “hitchhiking” feature affects all terms that share genes, inflating their significance, often giving them better statistical significance than other relevant terms, therefore misleading the interpretation of the results. This effect also produces very repetitive results, with many pathways that can have more than 90% of genes in common.

PathwayDenester takes the other pathways into consideration and evaluates each pathway based on the likelihood that it was enriched by its own and not by the genes in the intersection with more significant pathways. This is achieved by testing each pathway A with intersecting pathway B, taking into account the size of pathway A, its number of selected genes, the number of genes in common with pathway B, and the number of selected genes in their intersection. If the amount of selected genes that would fall in the intersection with pathway B is smaller than expected by chance, then PathwayDenester considers that the less significant pathway has its enrichment dependent on the more significant pathway and therefore should not be considered a good representative of the dataset.

When benchmarked with real and simulated datasets, PathwayDenester filtered out between ~40% and ~90% of imputed pathways, making the enriched pathway list objectively more concise and relevant, without loss of important biological insights.

PathwayDenester provides a systematic approach to reduce redundancy and dependency among enriched pathways, improving the interpretability and biological relevance of pathway enrichment results. PathwayDenester is freely available as an open-source Python command line program and as a web-service through https://pathwaydenester.pythonanywhere.com/