Modelling Self-Organization in Complex Networks Via a Brain-Inspired Network Automata Theory Improves Link Reliability in Protein Interactomes

Carlo Vittorio CannistraciScientific 2018 · DOI

TL;DR

This paper introduces a novel framework, the Local-Community-Paradigm (LCP), to predict the reliability of protein interactions within complex biological networks. By modeling self-organization principles inspired by brain networks, the authors propose four new models, with "C1*" emerging as the top performer, demonstrating that isolated local communities and low external interaction propensity are key indicators of reliable protein links.

What Problem Does This Paper Address?

Protein-protein interaction networks (PPINs), or interactomes, are notoriously "noisy" and incomplete. This means they contain many false positives (interactions that aren't real) and miss many true positives. This unreliability poses a significant challenge for downstream applications in systems biology and network medicine, such as predicting protein complexes, inferring disease pathways, or understanding gene function. Existing methods for assessing the reliability of these interactions often rely solely on network topology. However, the authors argue that while common neighborhood approaches are useful, they haven't fully captured the underlying self-organization principles that govern how these networks form and maintain their structure. This paper aims to bridge this gap by developing new topological models that better reflect these principles, specifically focusing on the concept of "local communities" within the network.

Methods

The core of this work lies in extending the Local-Community-Paradigm (LCP) theory. The LCP theory posits that network growth and organization are influenced by local topological structures, particularly "local communities" formed by common neighbors of interacting nodes. The authors introduce four new LCP-based models (C1, C1, C2, C2) that mathematically formalize two key rules for link reliability:

  1. Isolation of common neighbors: The reliability of an interaction is higher if the common neighbors of the interacting proteins are themselves isolated within a local community, meaning they have few connections to proteins outside this community.
  2. Low external propensity: The reliability is also higher if the interacting proteins have a low tendency to form links with proteins outside their local community.

These rules are combined in different ways within the proposed models. Notably, C1* mathematically integrates these two rules, with a specific emphasis on minimizing external links. The authors also introduce a revised evaluation framework using Gene Ontology (GO) semantic similarity, focusing on the intersection of biological process (BP) and cellular component (CC) categories for a more stringent assessment of true interactions.

Datasets

The study utilizes nine high-quality network datasets from various organisms, including:

  • Yeast (S. cerevisiae): 3 datasets
  • Plant (A. thaliana): 1 dataset
  • Worm (C. elegans): 1 dataset
  • Fly (D. melanogaster): 1 dataset
  • Human (H. sapiens): 4 datasets, including one particularly high-quality and structurally resolved network.

The paper also introduces a refined GO-based evaluation framework, which is presented as a more conservative and stringent benchmark compared to previous studies.

Key Insights and Ideas

  • Local communities matter: The paper strongly suggests that the organization of proteins into "local communities" is a fundamental principle driving the reliability of interactions. Proteins within these isolated groups are more likely to interact reliably.
  • Minimizing external links is crucial: The most significant finding is that a low propensity for proteins to interact with those outside their local community is a powerful indicator of reliable interactions. This is captured by the C1* model.
  • Brain inspiration yields results: The LCP theory, inspired by brain network self-organization, proves effective in understanding and modeling complex biological networks like protein interactomes.
  • Normalization improves performance: Normalized topological indices consistently outperform their non-normalized counterparts, highlighting the importance of considering the overall network context rather than just raw counts of neighbors.
  • MF is a weak indicator for reliability: The study suggests that using Molecular Function (MF) GO annotations alone for evaluating protein interaction reliability might be misleading, while Biological Process (BP) and Cellular Component (CC) are more robust indicators.

Key Results

The proposed C1* model consistently outperformed all other tested methods across various networks and evaluation metrics.

  • Top Performer: C1* was the best-performing algorithm in 6 out of 9 networks and ranked among the top positions in the remaining three.
  • Statistical Significance: The performance difference between C1* and other methods was statistically significant (Mann-Whitney test) for both Area Under Precision (AUP) and scaled Area Under Recall (sAUR).
  • Robustness: C1* demonstrated robust performance across different network sizes and organisms.
  • Normalized vs. Non-Normalized: Normalized algorithms generally showed significantly better performance than non-normalized ones.
  • C1's Rule: The rule captured by C1 (minimizing external links of common neighbors) was found to be more important than the rule captured by CN (common neighbors) for protein interaction reliability.

Limitations and Future Work

  • Network Incompleteness: The authors acknowledge that protein interaction networks are inherently incomplete and subject to "research bias" (e.g., well-studied proteins having more connections). This incompleteness can influence the performance of topological methods.
  • Future Work on Perturbations: Future studies could investigate how topological perturbations (e.g., edge removal, rewiring, adding false positives) affect the performance of these methods.
  • Generative Models: The use of generative models to simulate network properties and benchmark methods is suggested, though the challenge of identifying the "authentic" generative model for protein networks remains.
  • Comparison with Global Methods: The authors plan to compare their novel local models with state-of-the-art global methods (like machine learning) in future work to identify complementary strengths.
  • Merging Predictors: Exploring strategies to merge predictions from different types of predictors (local and global) into a single tool is suggested for enhanced reliability assessment.
  • Denominator Exponent: While C1 with an exponent of 2 performed best, further exploration of different denominator exponents for C1 and C2* is suggested.

Summary updated

Abstract

Protein interactomes are epitomes of incomplete and noisy networks. Methods for assessing link-reliability using exclusively topology are valuable in network biology, and their investigation facilitates the general understanding of topological mechanisms and models to draw and correct complex network connectivity. Here, I revise and extend the local-community-paradigm (LCP). Initially detected in brain-network topological self-organization and afterward generalized to any complex network, the LCP is a theory to model local-topology-dependent link-growth in complex networks using network automata. Four novel LCP-models are compared versus baseline local-topology-models. It emerges that the reliability of an interaction between two proteins is higher: (i) if their common neighbours are isolated in a complex (local-community) that has low tendency to interact with other external proteins; (ii) if they have a low propensity to link with other proteins external to the local-community. These two rules are mathematically combined in C1*: a proposed mechanistic model that, in fact, outperforms the others. This theoretical study elucidates basic topological rules behind self-organization principia of protein interactomes and offers the conceptual basis to extend this theory to any class of complex networks. The link-reliability improvement, based on the mere topology, can impact many applied domains such as systems biology and network medicine.
BibTeX
@inproceedings{cannistraci2018modelling,
  author = {Carlo Vittorio Cannistraci},
  title = {{Modelling Self-Organization in Complex Networks Via a Brain-Inspired Network Automata Theory Improves Link Reliability in Protein Interactomes}},
  booktitle = {Scientific},
  year = {2018},
  doi = {10.1038/s41598-018-33576-8},
}
Figure