Identifying strengths and weaknesses of methods for computational network inference from single-cell RNA-seq data.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables deep analysis of cellular transcriptional states.
• Inferring gene regulatory networks from scRNA-seq data is a critical but challenging task.
• Numerous computational methods exist, necessitating robust benchmarking.

Purpose of the Study:

• To comprehensively benchmark eleven recent gene regulatory network inference methods using scRNA-seq data.
• To evaluate methods based on computational demands and network recovery accuracy.
• To compare scRNA-seq derived networks against bulk data networks.

Main Methods:

• Utilized seven diverse scRNA-seq datasets from human, mouse, and yeast.
• Assessed eleven network inference algorithms, considering various gold standards and metrics.
• Evaluated performance based on Area Under the Precision Recall curve and biological relevance.
• Compared methods using only expression data versus those incorporating prior biological knowledge.

Main Results:

• Most methods showed modest recovery of known interactions globally.
• Methods effectively captured biologically relevant regulatory targets.
• Top-performing methods using only expression data included SCENIC, PIDC, MERLIN, and Correlation.
• Methods integrating prior biological knowledge (Inferelator, MERLIN) outperformed expression-only methods.
• Data imputation did not enhance, and sometimes harmed, network inference accuracy.
• Networks inferred from scRNA-seq data were comparable or superior to those from bulk RNA-seq data.

Conclusions:

• Prior biological knowledge significantly improves gene regulatory network inference from scRNA-seq.
• Current methods show promise but require further development and better gold standards.
• scRNA-seq is a viable and powerful data source for inferring gene regulatory networks.