Seiðr: Efficient calculation of robust ensemble gene networks

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Gene regulatory and co-expression networks are vital for analyzing high-dimensional gene expression data.
• Existing methods often suffer from low signal-to-noise ratios, non-linear interactions, and dataset biases.
• Aggregating networks from multiple methods can yield improved, more robust results.

Purpose of the Study:

• To introduce Seidr, a scalable software toolkit for gene regulatory and co-expression network inference.
• To address limitations of existing network inference methods, including algorithmic bias and noisy edges.
• To provide a practical tool for scientists to perform best-practice network analyses.

Main Methods:

• Seidr constructs community networks to mitigate algorithmic bias.
• It employs noise-corrected network backboning to refine network edges.
• The toolkit was benchmarked across model organisms: Saccharomyces cerevisiae, Drosophila melanogaster, and Arabidopsis thaliana.

Main Results:

• Individual network inference algorithms exhibit biases towards specific gene-gene interactions.
• Seidr's community network approach demonstrates reduced bias and robust performance across diverse benchmarks.
• The software successfully identified key components and suggested gene functions in a Norway spruce drought stress network.

Conclusions:

• Seidr offers a robust and less biased approach to gene network inference compared to individual algorithms.
• The community network strategy effectively reduces algorithmic bias and improves reliability.
• Seidr is a valuable tool for analyzing gene expression data, even in non-model species, aiding in functional gene discovery.