A best-match approach for gene set analyses in embedding spaces.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Embedding methods reduce high-dimensional biological data into lower-dimensional spaces.
• Gene embeddings capture relationships between genes but are mainly used for machine learning.
• Direct analysis of gene sets within embedding spaces is underexplored.

Purpose of the Study:

• Introduce a novel algorithm, ANDES (Algorithm for Network Data Embedding and Similarity), for analyzing gene sets in embedding spaces.
• Demonstrate ANDES's utility in comparing gene sets while accounting for diversity.
• Explore ANDES's potential for cross-organism functional knowledge transfer.

Main Methods:

• ANDES is a best-match approach applied to existing gene embeddings.
• It is used for gene set enrichment analysis (overrepresentation and rank-based).
• The method integrates multiorganism joint gene embeddings for cross-species comparisons.

Main Results:

• ANDES achieves state-of-the-art performance in gene set enrichment analysis.
• It effectively reconciles gene set diversity for improved comparisons.
• ANDES facilitates phenotype mapping across different model organisms.

Conclusions:

• ANDES offers a flexible and intuitive method for analyzing gene sets in embedding spaces.
• It enhances the utility of embeddings for biological data interpretation.
• The methodology can be extended to other embedding spaces with complex community structures.