Uncovering latent biological function associations through gene set embeddings

TL;DR

This paper introduces a new method using gene set embeddings to uncover biological relationships and associations across species.

Contribution

The novel contribution is integrating attribute-driven knowledge with network analysis to reveal both expected and new biological insights.

Findings

The framework validates cross-species associations using human and mouse data in a shared vector space.

It uncovers potential gene-biological term associations through network connectivity analysis.

The method complements traditional biological network analysis with a comprehensive perspective.

Abstract

The complexity of biological systems has increasingly been unraveled through computational methods, with biological network analysis now focusing on the construction and exploration of well-defined interaction networks. Traditional graph-theoretical approaches have been instrumental in mapping key biological processes using high-confidence interaction data. However, these methods often struggle with incomplete or/and heterogeneous datasets. In this study, we extend beyond conventional bipartite models by integrating attribute-driven knowledge from the Molecular Signatures Database (MSigDB) using the node2vec algorithm.

Our approach explores unsupervised biological relationships and uncovers potential associations between genes and biological terms through network connectivity analysis. By embedding both human and mouse data into a shared vector space, we validate our findings cross-species, further strengthening the robustness of our method.

This integrative framework reveals both expected and novel biological insights, offering a comprehensive perspective that complements traditional biological network analysis and paves the way for deeper understanding of complex biological processes and diseases.

The online version contains supplementary material available at 10.1186/s12859-025-06100-9.

Figures