A novel probabilistic generator for large-scale gene association networks.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Gene expression data allows for inferring gene-gene associations using network inference methods.
• Assessing network inference performance is challenging due to unknown true networks in real data.
• Current benchmarks using reference networks fail to capture topological heterogeneity across organisms and tissues.

Purpose of the Study:

• To develop a novel probabilistic network generator for robust benchmarking of network inference methods.
• To address limitations of reference-based generators by creating realistic gene association networks.
• To capture the heterogeneity of network topologies across different organisms and tissues.

Main Methods:

• Proposed a novel probabilistic network generator.
• Generated organism-specific and human tissue-specific gene association networks.
• Utilized global topology measures to assess similarity between generated and gold-standard networks.

Main Results:

• The proposed generator effectively creates realistic gene association networks.
• Demonstrated significant variability in network structures across different organisms and tissues.
• Showed that the commonly used "scale-free" model is inadequate for replicating biological network structures.

Conclusions:

• Probabilistic generators offer a robust approach for benchmarking network inference methods.
• The novel generator captures biological network heterogeneity better than existing methods.
• Results highlight the inadequacy of the "scale-free" model for biological networks.