Netter: re-ranking gene network inference predictions using structural network properties.

Area of Science:

• Bioinformatics
• Systems Biology
• Computational Biology

Background:

• Gene regulatory network (GRN) inference from expression data is crucial for understanding cellular mechanisms.
• Existing GRN inference algorithms often yield topologies with suboptimal structural properties, limiting their practical accuracy.
• Current methods primarily rely on data correlations, neglecting known biological network characteristics.

Purpose of the Study:

• To introduce Netter, a novel post-processing algorithm for refining GRN predictions.
• To enhance the accuracy of inferred network topologies by integrating graph properties.
• To provide a flexible tool applicable to diverse network inference methods and omics data.

Main Methods:

• Developed a post-processing algorithm (Netter) to re-rank gene interaction confidence scores.
• Utilized graphlets and graph-invariant properties as optimization criteria for re-ranking.
• Applied Netter to re-rank predictions from six state-of-the-art network inference algorithms.

Main Results:

• Netter significantly improved predictions on both synthetic and benchmark datasets (DREAM4, DREAM5).
• The algorithm enhanced the GRN inference for E. coli using real expression data.
• Netter demonstrated robustness across various parameter settings and outperformed other post-processing methods.

Conclusions:

• Netter offers a flexible and effective second step to improve GRN inference quality.
• The algorithm can be customized with user-defined graph properties for specific prior knowledge.
• Netter enhances the utility of network inference methods by refining topological predictions.