Improving gene regulatory network inference and assessment: The importance of using network structure

Area of Science:

• Bioinformatics
• Systems Biology
• Computational Biology

Background:

• Gene regulatory networks (GRNs) model transcription, but are incomplete due to experimental costs.
• Current GRN inference methods show modest performance based on gene expression data.
• Assessing these methods is challenging due to data quality and gold standard limitations.

Purpose of the Study:

• To investigate caveats in GRN inference and assessment.
• To evaluate the impact of input data quality and gold standard on network inference.
• To focus on global network structure for method assessment.

Main Methods:

• Utilized synthetic and biological gene expression data for predictions.
• Employed experimentally-validated biological networks as the gold standard (ground truth).
• Analyzed standard performance metrics and graph structural properties.

Main Results:

• Methods inferring regulatory interactions outperform co-expression methods for global GRN inference.
• Co-expression methods are more suitable for function-specific regulons and co-regulation networks.
• Merging expression data requires balancing data size against noise and considering graph structure.

Conclusions:

• Regulatory and co-expression network inference methods should not be assessed using the same criteria.
• Guidelines are provided for leveraging inference methods and their assessment based on applications and datasets.
• Future assessments must account for network structure and data integration challenges.