Inferring functional communities from partially observed biological networks exploiting geometric topology and side information

Jayson Sia, Wei Zhang, Edmond Jonckheere, David E. Cook, Paul BogdanScientific 2022 · DOI

Abstract

Cellular biological networks represent the molecular interactions that shape function of living cells. Uncovering the organization of a biological network requires efficient and accurate algorithms to determine the components, termed communities, underlying specific processes. Detecting functional communities is challenging because reconstructed biological networks are always incomplete due to technical bias and biological complexity, and the evaluation of putative communities is further complicated by a lack of known ground truth. To address these challenges, we developed a geometric-based detection framework based on Ollivier-Ricci curvature to exploit information about network topology to perform community detection from partially observed biological networks. We further improved this approach by integrating knowledge of gene function, termed side information, into the Ollivier-Ricci curvature algorithm to aid in community detection. This approach identified essential conserved and varied biological communities from partially observed Arabidopsis protein interaction datasets better than the previously used methods. We show that Ollivier-Ricci curvature with side information identified an expanded auxin community to include an important protein stability complex, the Cop9 signalosome, consistent with previous reported links to auxin response and root development. The results show that community detection based on Ollivier-Ricci curvature with side information can uncover novel components and novel communities in biological networks, providing novel insight into the organization and function of complex networks.

Introduction

Biological networks, such as protein-protein interaction (PPI) maps, serve as the blueprints for cellular function. However, these maps are notoriously noisy and incomplete due to the inherent limitations of experimental sampling and biological complexity. Traditional community detection algorithms, including modularity-based methods and the Girvan-Newman algorithm, often assume that the provided network topology is complete. In reality, researchers frequently deal with "partially observed" networks where many interactions remain hidden.

Furthermore, while researchers often possess "side information" (SI)—prior knowledge regarding gene functions or metabolic pathways—this information is typically relegated to post-hoc validation. This paper introduces a framework that integrates SI directly into the topological analysis to improve the accuracy of community detection in fragmented biological datasets.

Methodology

The core of the proposed approach is ORCCI (Ollivier-Ricci Curvature Community Identification), a framework that treats network edges as geometric objects to identify functional modules.

Ollivier-Ricci Curvature (ORC)

ORC provides a geometric measure of the local connectivity of a graph. In this context, edges with positive curvature indicate regions of high local clustering, characteristic of functional communities. Conversely, edges with negative curvature represent "bridges" or bottlenecks that connect disparate communities. The standard ORC algorithm identifies communities by iteratively removing edges with the most negative curvature, effectively pruning the network into its constituent modules.

Integration of Side Information

To handle incomplete data, the authors modified the edge-removal step. If a candidate edge for removal connects two nodes that share the same functional label (the side information), the edge is retained. This constraint prevents the algorithm from over-splitting known functional groups when the network structure is sparse, using prior knowledge as a structural stabilizer.

The ORCCI workflow: from a complete (unobserved) network through data collection to community detection using topology and side information.

Experimental Datasets

The framework was validated across synthetic and real-world biological networks:

  • Synthetic Networks: Generated using the Stochastic Block Model (SBM) with 1,000 nodes and 10 ground-truth communities to test varying levels of network observability (20% to 100%).
  • Arabidopsis thaliana PPI Networks: The study utilized curated sub-networks (PhI_MAIN, PhI_FULL) and a genome-wide interactome (AI-1_MAIN) containing over 2,600 proteins.

Results and Discussion

Performance Gains from Side Information

Synthetic tests revealed that side information is most impactful when network observability is between 40% and 80%. In these "gray zones," SI acts as a bridge that maintains the integrity of functional modules that would otherwise appear disconnected due to missing edges. As shown in the performance metrics, the integration of SI significantly boosts the Normalized Mutual Information (NMI) scores compared to topology-only methods.

Impact of side information on community detection accuracy (NMI) across different levels of network observability.

Biological Insights: The Auxin-Cop9 Connection

A significant discovery using the ORCCI+SI method was the identification of an expanded auxin response community. The algorithm successfully linked the core auxin pathway to the Cop9 signalosome (CSN5A subunit). While previous literature had suggested biological links between these two, the ORCCI framework provided a structural network basis for this crosstalk, demonstrating how a hub protein can bridge distinct functional sub-communities.

Accuracy in Real-World Networks

On the Arabidopsis PhI_MAIN network, ORC-based methods outperformed standard edge-betweenness baselines. Specifically, the ORC-MM (maximum modularity) partition achieved a 70% average pathway protein ratio, indicating that the detected communities were highly enriched for proteins sharing the same functional labels.

Conclusion and Future Directions

The ORCCI framework demonstrates that geometric topology is a powerful indicator of biological organization. By incorporating side information as a structural guide, the method overcomes the limitations of partially observed data. However, the authors note that the quality of detection is sensitive to the accuracy of the SI provided. Future work will focus on non-deterministic uses of side information, such as edge weighting, and utilizing SI to actively reconstruct missing portions of the interactome. While computationally intensive, this geometric approach provides a robust foundation for uncovering novel components in complex biological networks.

Summary updated

Abstract

Cellular biological networks represent the molecular interactions that shape function of living cells. Uncovering the organization of a biological network requires efficient and accurate algorithms to determine the components, termed communities, underlying specific processes. Detecting functional communities is challenging because reconstructed biological networks are always incomplete due to technical bias and biological complexity, and the evaluation of putative communities is further complicated by a lack of known ground truth. To address these challenges, we developed a geometric-based detection framework based on Ollivier-Ricci curvature to exploit information about network topology to perform community detection from partially observed biological networks. We further improved this approach by integrating knowledge of gene function, termed side information, into the Ollivier-Ricci curvature algorithm to aid in community detection. This approach identified essential conserved and varied biological communities from partially observed Arabidopsis protein interaction datasets better than the previously used methods. We show that Ollivier-Ricci curvature with side information identified an expanded auxin community to include an important protein stability complex, the Cop9 signalosome, consistent with previous reported links to auxin response and root development. The results show that community detection based on Ollivier-Ricci curvature with side information can uncover novel components and novel communities in biological networks, providing novel insight into the organization and function of complex networks.
BibTeX
@inproceedings{sia2022inferring,
  author = {Jayson Sia and Wei Zhang and Edmond Jonckheere and David E. Cook and Paul Bogdan},
  title = {{Inferring functional communities from partially observed biological networks exploiting geometric topology and side information}},
  booktitle = {Scientific},
  year = {2022},
  doi = {10.1038/s41598-022-14631-x},
}
Figure