TL;DR
This paper introduces MOPHY, a novel phylogenetic framework that analyzes the evolutionary history of protein interaction networks by focusing on conserved network modules rather than direct topological comparisons. This approach proves more robust to the inherent noise and incompleteness of biological data, offering a more accurate reconstruction of evolutionary relationships between species.
What Problem Does This Paper Address?
Understanding the evolution of biological systems is a fundamental goal in molecular biology. While traditional phylogenetic analyses rely on sequence data, this paper argues that the structure and function of molecular networks, particularly protein-protein interaction (PPI) networks, also carry crucial evolutionary information. However, comparing these networks directly is challenging due to their incompleteness, noise, and the computational intractability of subgraph isomorphism. Existing methods often struggle with these data imperfections. This paper aims to bridge this gap by proposing a method that leverages the concept of "network modules" – groups of proteins that perform a specific function and tend to be conserved together – to infer evolutionary relationships.
Methods
The core of the MOPHY framework involves several key steps:
- Module Identification: For each individual PPI network, MOPHY identifies functional and topological modules. This is achieved using a hierarchical clustering algorithm that integrates protein proximity (based on reliable paths) and mutual clustering coefficients to find tightly coupled subgraphs with low diameter.
- Module Projection: Identified modules from one network are then projected onto the networks of other species. This projection is based on the conservation of network proximity between homologous proteins. Essentially, it assesses how well the functional association of a module's components is maintained in other species.
- Module Map Construction: For each species, a "module map" is created. This map is a feature vector where each element represents the modularity score of a specific module in that species' network. These maps serve as high-level signatures of each network.
- Phylogenetic Tree Reconstruction: Pairwise distances between species are calculated by comparing their module maps. These distances are then used to construct a phylogenetic tree using a standard neighbor-joining algorithm.
A key innovation is the use of network proximity and modularity as features, rather than direct topological comparisons, which helps overcome the challenges of noisy and incomplete data.
Datasets
The paper evaluates its method using two types of datasets:
- Simulated Networks: The authors generated synthetic PPI networks by simulating a network evolution process (a modified duplication-mutation-complementation model with speciation events). This allowed them to create networks with a known underlying phylogeny, enabling a direct comparison of reconstructed trees with the true evolutionary history. These simulations were used to assess accuracy, robustness to noise, and the impact of various parameters (coverage, diameter).
- Extant PPI Networks: The method was also tested on real-world PPI data from seven diverse species: D. melanogaster, S. cerevisiae, E. coli, C. elegans, H. sapiens, H. pylori, and M. musculus. This data was obtained from the Database of Interacting Proteins (DIP).
Key Insights and Ideas
- Modularity as an Evolutionary Signal: The paper strongly advocates for the idea that conserved network modules, rather than individual interactions or overall network topology, are powerful indicators of evolutionary relationships.
- Robustness to Noise and Incompleteness: By focusing on aggregate network properties (modularity and proximity) rather than exact matches of individual interactions, MOPHY demonstrates significant robustness to the inherent noise and missing data in PPI networks. This is a critical advantage for biological data.
- Beyond Sequence Similarity: The method effectively integrates sequence homology information but crucially goes beyond it by incorporating network structure. This allows it to capture functional evolution that might not be apparent from sequence data alone.
- Evolutionary Trajectories of Modules: The framework opens avenues for studying the evolutionary histories of specific modules, potentially leading to a "periodic table of systems biology."
Key Results
- Superior Performance on Simulated Data: MOPHY consistently outperformed other methods (RDL, random modules, protein similarity only, random homolog selection) in reconstructing both the topology and evolutionary distances of simulated phylogenetic trees. For instance, MOPHY achieved significantly lower nodal distances compared to randomized methods.
- Robustness to Noise: Even with 50% noise (swapped interactions) or 40% correlated missing data, MOPHY's accuracy remained statistically significant, outperforming randomized methods. Uncorrelated missing data had a more detrimental effect, as expected.
- Accurate Reconstruction of Extant PPI Networks: When applied to real PPI data from seven species, MOPHY reconstructed a phylogenetic tree that aligned well with known phylogenetic relationships and evolutionary distances inferred by independent methods. This tree was also consistent with a whole-genome based phylogeny.
- Module Specificity Matters: The study found that using "most specific" (smaller, tightly coupled) modules generally led to better performance than using "most comprehensive" (larger) modules, especially when considering evolutionary distances.
Limitations and Future Work
- Network Distance Overestimation: The paper notes that network-based distance measures can tend to overestimate evolutionary distances between species. Methods for normalizing these distances are suggested as necessary.
- Impact of Data Availability: While MOPHY is robust, the paper acknowledges that significant differences in network size due to varying data availability can still introduce artificial variation. The method's ability to handle this by focusing on conserved modules is highlighted as a strength.
- Future Work: The authors suggest further elaborate studies on modular evolution, including the identification of module families and the reconstruction of their evolutionary trajectories. This could lead to a more comprehensive understanding of functional module evolution. They also propose that the framework can be extended to other types of biological network data (e.g., gene expression, regulatory networks).
Summary updated