JEBIN: analyzing gene co-expressions across multiple datasets by joint network embedding.

Area of Science:

• Bioinformatics and Computational Biology
• Genomics and Transcriptomics
• Cancer Research

Background:

• Inferring gene co-expression associations is crucial for transcriptomic data analysis.
• High dimensionality and noise in single datasets hinder stable inference of gene co-expression.
• Meta-analysis of multisource data offers a robust solution for overcoming these limitations.

Purpose of the Study:

• To develop a novel method, Joint Embedding of multiple BIpartite Networks (JEBIN), for integrating multiple gene expression datasets.
• To learn a low-dimensional consensus representation of genes for accurate co-expression inference.
• To apply JEBIN to identify gene co-expression patterns in hepatocellular carcinoma (HCC) and discover potential therapeutic targets.

Main Methods:

• Proposed Joint Embedding of multiple BIpartite Networks (JEBIN) for integrating diverse transcriptomic datasets.
• JEBIN employs nonlinear and global similarity for inferring gene co-expression associations.
• The method demonstrates linear time complexity concerning gene and total sample size, enabling scalability.

Main Results:

• JEBIN's effectiveness and scalability were validated through simulation experiments.
• Comparative analysis on real biological datasets showed JEBIN's superiority over existing integration methods.
• Application to HCC data revealed significant differences in gene co-expression between bulk and single-cell datasets.
• Numerous differentially co-expressed ligand-receptor pairs were identified in HCC compared to adjacent normal tissues.

Conclusions:

• JEBIN provides a powerful and scalable approach for meta-analysis of transcriptomic data to infer gene co-expression.
• The identified differentially co-expressed ligand-receptor pairs offer promising candidates for targeted HCC therapies.
• The study highlights the distinct gene co-expression landscapes in bulk versus single-cell HCC data.