Benchmarking Gene Regulatory Network Inference Methods on Simulated and Experimental Data.

Area of Science:

• Computational Biology
• Systems Biology
• Genomics

Background:

• Gene regulatory network inference aims to elucidate gene interactions but its performance on real-world data remains uncertain.
• Previous benchmarking efforts on experimental data have produced inconsistent results, with methods sometimes failing to outperform random chance.
• There is a critical need for rigorous benchmarking of network inference methods using experimental data with known ground truth.

Approach:

• This study presents the first known benchmarking of gene regulatory network inference methods using single-cell transcriptomic data from E. coli.
• Methods were evaluated on their accuracy in reconstructing the known gene regulatory network.
• Computer simulations were employed to assess a simple network inference method (Pearson correlation) and explore best practices.

Key Points:

• Network inference methods demonstrated moderate accuracy on single-cell E. coli transcriptomic data, outperforming random guessing but falling short of perfection.
• Several methods that performed well on microarray and bulk RNA-sequencing data showed reduced accuracy when applied to single-cell data.
• Simulations indicated that using a simplified gene expression model omitting the mRNA step can significantly overestimate method accuracy.

Conclusions:

• Accurate benchmarking of gene regulatory network inference methods is crucial for understanding their real-world applicability.
• The transition to single-cell data presents unique challenges for network inference, requiring re-evaluation of existing methods.
• Future high-throughput proteomic data may offer improved accuracy for network inference compared to transcriptomic data.