Benchmarking algorithms for gene regulatory network inference from single-cell transcriptomic data.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Inferring gene regulatory networks (GRNs) from single-cell transcriptional data is crucial for understanding cellular processes.
• Existing algorithms face challenges in accuracy and validation due to data complexity and lack of ground truth.
• Previous evaluation methods for GRN inference algorithms have limitations.

Purpose of the Study:

• To systematically evaluate state-of-the-art algorithms for GRN inference from single-cell data.
• To develop a robust evaluation framework (BEELINE) for assessing algorithm performance.
• To provide recommendations for end-users selecting GRN inference tools.

Main Methods:

• Developed a novel strategy to simulate single-cell transcriptional data from synthetic and Boolean networks.
• Utilized synthetic networks, literature-curated Boolean models, and diverse experimental single-cell RNA-seq datasets for ground truth.
• Implemented the BEELINE evaluation framework to assess algorithm accuracy using metrics like area under the precision-recall curve and early precision.

Main Results:

• Most evaluated GRN inference algorithms demonstrated moderate performance in terms of area under the precision-recall curve and early precision.
• Algorithms performed better in recovering interactions in synthetic networks compared to Boolean models.
• Techniques that do not rely on pseudotime-ordered cells generally exhibited higher accuracy.
• Algorithms showing high early precision on Boolean models also performed well on experimental datasets.

Conclusions:

• Current state-of-the-art GRN inference algorithms have limitations in accuracy when applied to single-cell data.
• The BEELINE framework provides a standardized approach for evaluating and comparing GRN inference methods.
• Recommendations are provided to guide users in selecting appropriate algorithms based on data type and desired accuracy metrics.