Limits...
Evolution of protein-protein interaction networks in yeast

View Article: PubMed Central - PubMed

ABSTRACT

Interest in the evolution of protein-protein and genetic interaction networks has been rising in recent years, but the lack of large-scale high quality comparative datasets has acted as a barrier. Here, we carried out a comparative analysis of computationally predicted protein-protein interaction (PPI) networks from five closely related yeast species. We used the Protein-protein Interaction Prediction Engine (PIPE), which uses a database of known interactions to make sequence-based PPI predictions, to generate high quality predicted interactomes. Simulated proteomes and corresponding PPI networks were used to provide expectations for the extent and nature of PPI network evolution. We found strong evidence for conservation of PPIs, with lower than expected levels of change in PPIs for about a quarter of the proteome. Furthermore, we found that changes in predicted PPI networks are poorly predicted by sequence divergence. Our analyses identified a number of functional classes experiencing fewer PPI changes than expected, suggestive of purifying selection on PPIs. Our results demonstrate the added benefit of considering predicted PPI networks when studying the evolution of closely related organisms.

No MeSH data available.


An overview of the computational process used to infer PPI networks for                        each of the 5 yeast species, and to generate the simulated                         model.                    Molecular evolutionary parameters were inferred under the M0 model in PAML,                        and were used to generate simulated datasets using INDELible. PIPE was used                        to infer PPI networks for both the real and simulated datasets.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC5382968&req=5

pone.0171920.g001: An overview of the computational process used to infer PPI networks for each of the 5 yeast species, and to generate the simulated model. Molecular evolutionary parameters were inferred under the M0 model in PAML, and were used to generate simulated datasets using INDELible. PIPE was used to infer PPI networks for both the real and simulated datasets.

Mentions: We propose that two key problems—potential bias associated with using the PIPE algorithm for cross-species predictions, and the challenge of formulating a hypothesis for network evolution—can be addressed using a common approach. In both cases, expectations must be formulated with respect to the effects of mutations on the inferred PPI network: in the case of controlling for bias associated with PIPE, how do random mutations affect PIPE’s inferences? And, with respect to a hypothesis for network evolution, how much change in a PPI network is expected given random mutation (i.e., mutations that are random with respect to PPIs)? We provide expectations for changes in the inferred PPI network using simulated proteomes from the four non-cerevisiae yeasts. In the simulated proteomes, the locations of substitutions (both point mutations and insertion-deletions) are random with respect to PPIs. This is equivalent to assuming that natural selection does not operate for or against mutations that modify PPIs, as is appropriate in a model. The rates and types of substitutions, however, are modeled on the real sequence data. As such, the simulated datasets provide a baseline expectation for how many changes we expect to infer using the PIPE algorithm, given mutation but no natural selection on PPIs. Using simulations, we propose to both mitigate bias associated with PIPE, as well as to provide a hypothesis against which selection can be inferred. An overview of this process is illustrated in Fig 1.


Evolution of protein-protein interaction networks in yeast
An overview of the computational process used to infer PPI networks for                        each of the 5 yeast species, and to generate the simulated                         model.                    Molecular evolutionary parameters were inferred under the M0 model in PAML,                        and were used to generate simulated datasets using INDELible. PIPE was used                        to infer PPI networks for both the real and simulated datasets.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC5382968&req=5

pone.0171920.g001: An overview of the computational process used to infer PPI networks for each of the 5 yeast species, and to generate the simulated model. Molecular evolutionary parameters were inferred under the M0 model in PAML, and were used to generate simulated datasets using INDELible. PIPE was used to infer PPI networks for both the real and simulated datasets.
Mentions: We propose that two key problems—potential bias associated with using the PIPE algorithm for cross-species predictions, and the challenge of formulating a hypothesis for network evolution—can be addressed using a common approach. In both cases, expectations must be formulated with respect to the effects of mutations on the inferred PPI network: in the case of controlling for bias associated with PIPE, how do random mutations affect PIPE’s inferences? And, with respect to a hypothesis for network evolution, how much change in a PPI network is expected given random mutation (i.e., mutations that are random with respect to PPIs)? We provide expectations for changes in the inferred PPI network using simulated proteomes from the four non-cerevisiae yeasts. In the simulated proteomes, the locations of substitutions (both point mutations and insertion-deletions) are random with respect to PPIs. This is equivalent to assuming that natural selection does not operate for or against mutations that modify PPIs, as is appropriate in a model. The rates and types of substitutions, however, are modeled on the real sequence data. As such, the simulated datasets provide a baseline expectation for how many changes we expect to infer using the PIPE algorithm, given mutation but no natural selection on PPIs. Using simulations, we propose to both mitigate bias associated with PIPE, as well as to provide a hypothesis against which selection can be inferred. An overview of this process is illustrated in Fig 1.

View Article: PubMed Central - PubMed

ABSTRACT

Interest in the evolution of protein-protein and genetic interaction networks has been rising in recent years, but the lack of large-scale high quality comparative datasets has acted as a barrier. Here, we carried out a comparative analysis of computationally predicted protein-protein interaction (PPI) networks from five closely related yeast species. We used the Protein-protein Interaction Prediction Engine (PIPE), which uses a database of known interactions to make sequence-based PPI predictions, to generate high quality predicted interactomes. Simulated proteomes and corresponding PPI networks were used to provide expectations for the extent and nature of PPI network evolution. We found strong evidence for conservation of PPIs, with lower than expected levels of change in PPIs for about a quarter of the proteome. Furthermore, we found that changes in predicted PPI networks are poorly predicted by sequence divergence. Our analyses identified a number of functional classes experiencing fewer PPI changes than expected, suggestive of purifying selection on PPIs. Our results demonstrate the added benefit of considering predicted PPI networks when studying the evolution of closely related organisms.

No MeSH data available.