Guidance for RNA-seq co-expression network construction and analysis: safety in numbers

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• RNA-sequencing (RNA-seq) co-expression network analysis is an emerging field with undefined best practices.
• Understanding factors influencing functional connectivity and network topology in RNA-seq data is crucial.

Purpose of the Study:

• To assess RNA-seq expression data to identify factors impacting co-expression network properties.
• To compare RNA-seq co-expression networks with those derived from microarrays.

Main Methods:

• Utilized a Guilt-By-Association framework on 1970 RNA-seq samples.
• Examined gene co-expression tendencies to infer shared function.
• Evaluated network performance using the area under the receiver operator characteristic curve.

Main Results:

• Minimal criteria for RNA-seq network performance comparable to microarrays include >20 samples with >10M reads/sample.
• Aggregate RNA-seq networks achieved an AUC of ~0.71.
• Thousands of samples are needed for 'gold-standard' co-expression, similar to microarrays.
• Significant topological differences exist between RNA-seq and microarray networks, particularly in hub gene overlap, due to technology-specific expression noise correlations.

Conclusions:

• RNA-seq co-expression network analysis requires substantial sample sizes for robust results.
• Differences in expression noise between RNA-seq and microarrays lead to distinct network topologies.
• Further research is needed to refine practices for RNA-seq co-expression network construction.