Juxtapose: a gene-embedding approach for comparing co-expression networks.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Comparing complex gene co-expression networks (GCNs) across species presents significant challenges.
• Existing methods struggle with the intricate structures of GCNs, limiting comparative transcriptomics.
• A novel tool, Juxtapose, is introduced to address these limitations.

Purpose of the Study:

• To develop and validate Juxtapose, a tool for comparing gene co-expression networks across different organisms.
• To establish similarity measures for effective comparative transcriptomics.
• To enable evolutionary studies and comparisons across tissues or conditions within species.

Main Methods:

• Utilized a word embedding strategy from natural language processing to generate gene embeddings from GCNs.
• Employed RNA-seq datasets from GTEx and Gene Expression Omnibus, alongside synthesized data for evaluation.
• Performed biological evaluation using gene set enrichment analysis and literature-based gene relationships.

Main Results:

• Juxtapose successfully aligned synthetic networks and identified conserved regions in real GCNs without external biological information.
• Cosine distance between GCN embeddings proved to be an informative measure of topological similarity.
• The tool demonstrated consistency in embedding nodes of synthetic network structures.

Conclusions:

• Juxtapose facilitates GCN alignment independent of known biological similarities.
• Enables post-hoc analyses leveraging biological parameters like gene orthology and pathway conservation.
• Provides a versatile platform for comparative transcriptomics in evolutionary and other biological studies.