Identifying large-scale interaction atlases using probabilistic graphs and external knowledge.

Area of Science:

• Systems Biology
• Bioinformatics
• Computational Biology

Background:

• Reconstructing gene interaction networks is crucial for understanding biological mechanisms but challenging due to data noise and network size.
• Complex models handle data stochasticity for small networks, while simpler models generate large but less reliable networks.

Purpose of the Study:

• To develop a robust and scalable method for reconstructing gene interaction networks from experimental data.
• To integrate external knowledge and probabilistic graph representations for improved interactome reconstruction.

Main Methods:

• A divide-and-conquer strategy was employed, involving data clustering and learning interaction networks for each cluster.
• Networks were merged using representative genes from each cluster, incorporating probabilistic graph representations and external knowledge.

Main Results:

• An interaction atlas of 11,454 genes and 17,777 edges was generated for 337 human pathways.
• The proposed approach demonstrated a significant performance improvement, outperforming baseline methods by approximately 5-17 fold based on precision-recall curves.
• Performance was strongly correlated with the accuracy of the clustering step in identifying biological modularity.

Conclusions:

• The developed workflow optimizes algorithm and parameter selection for interaction atlas generation.
• The approach effectively integrates external knowledge into interactome reconstruction using probabilistic graphs.
• Network characterization and analysis of long-range effects have implications for biomarker discovery and therapeutic strategies.