RNA-seq preprocessing and sample size considerations for gene network inference.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene network inference (GNI) methods identify gene functional relationships.
• Adapting GNI for RNA-seq data requires understanding optimal preprocessing.
• Sample size needs for reliable RNA-seq GNI are not well-defined.

Approach:

• Evaluated 300 preprocessing combinations across 7 RNA-seq datasets.
• Assessed data transformations, normalizations, and association estimators.
• Determined optimal preprocessing strategies and required sample sizes for GNI.

Key Points:

• No single preprocessing method is universally optimal for RNA-seq GNI.
• Log-2 TPM with VST and PCC, or raw counts with PCC, are promising general strategies.
• Performance varies by dataset, necessitating empirical evaluation.
• 30 to 85 samples are recommended for reliable GNI estimates.

Conclusions:

• Provides practical recommendations for RNA-seq data preprocessing for GNI.
• Highlights the importance of dataset-specific evaluation of preprocessing pipelines.
• Offers guidance on sample size for robust gene network inference.