Research On The Relationship Between Topological Structure And Functional Lethality In Protein-protein Interaction Network

With the success of Human Genome Project (HGP) and the accomplishment ofmodel organism sequencing, bioinformatics steps into the post-genome era. Along withthe development of high-throughput technologies and the acquirement of large amountsof biological data,‘X-omics’, such as genomics, transcriptomics, proteomics andmetabonomics, have become hot research areas gradually, and especially, give birth tosystems biology.Massive amounts of protein-protein interaction (PPI) data, which were producedby high-throughput technologies, provide abundant data sources for constructing PPInetworks and also make the research on PPI networks possible. As an importantcomponent of biological networks, PPI networks provide the foundation to study otherbiological networks. The exploration on PPI networks is of great help to systematicallyunderstand different kinds of biological processes and the law of life activities.In this thesis, we primarily focus on the fundamental problem of systems biology,i.e.‘the structure decides its functions’. Our work is designed to tackle this problem inPPI networks. Specifically, we explore the relationship between proteins’ topologicalcentrality and functional lethality. Firstly, we classify the proteins into hub proteins andnon-hub proteins according to their degrees, and then research the functional lethality ofhub proteins and non-hub proteins. The main contributions are summarized as follows.(1) In PPI networks, centrality of hub proteins is bound up with their lethality. Butso far, there is no identical interpretation of this relevance. And there are false hubproteins in PPI networks, which are high degrees without lethality. There’s noreasonable explanation why the ‘centrality-lethality’ rule dose not apply to false hubproteins. To address these problems, we compare false hub proteins with essential hubproteins on their topological attribute and evolutionary attribute in unweighted andweighted PPI networks, respectively. The experimental results show that, essential hubproteins and false hub proteins have significant differences in centrality measures,clustering coefficient and evolutionary rate. More specifically, essential hub proteinshave higher centrality measures, higher clustering coefficient and lower evolutionaryrate, while false hub proteins have lower centrality measures, lower clusteringcoefficient and higher evolutionary rate. The differences tell that, essential hub proteinshave higher topological centrality and evolutionary conservation, while false hubproteins have lower topological centrality and are more likely to evolve. At the sametime, these differences explain the ‘centrality-lethality’ rule well in terms of topologyand evolution.(2) The topology-based centrality measures are sensitive to noises in PPI networks.So the computational results are always inconsistent, even contrary to each other. In order to solve this problem, we combine both the topological and functional informationof proteins and propose four new measures: n-neighbor, n-iep, n-ipc and n-f. All thenew measures are robust to noises in datasets, and they reflect the innate characters ofproteins because of the combination of functional information. The experimental resultsshow that, compared to essential hub proteins, false hub proteins have less appearancefrequency in protein complexes, and lower n-neighbor, n-iep, n-ipc and n-f values,which explains the causation of the ‘centrality-lethality’ rule in terms of functions andwhy false hub proteins have no lethality.(3) Some researches show that hub proteins can be further classified into party hubproteins and date hub proteins. However, this kind of classification is under debate.Aiming at this problem, this thesis defines the concepts of E-party hub proteins andE-date hub proteins, the study on whose topology and function is helpful to settle thedispute. The experimental results show that, E-party hub proteins and E-date hubproteins have significant differences in centrality measures, clustering coefficient, cellcycle, stress response and DNA damage. The results support the further classification ofhub proteins, and also demonstrate that E-party hub proteins are tend to be intra-modulehub proteins while E-date hub proteins are tend to be inter-module hub proteins.(4) There are some essential proteins with low degrees (essential non-hub proteins)in PPI networks. This is another exception of the centrality-lethality rule. The reasonwhy these essential non-hub proteins are lethality is still unknown. In order to solve thisproblem, we compare essential non-hub proteins with non-essential non-hub proteinsfrom the functional view. The experimental results show that there are more essentialnon-hub proteins in protein complexes and they have higher n-neighbor, n-iep and n-ipcvalues. These results prove that essential non-hub proteins involve in more denseregions and they are tend to interact with essential proteins in PPI networks. Andneighbors of essential non-hub proteins are close, too. These are all reasons foressentiality of essential non-hub proteins.(5) The existing methods for prediction essential proteins only consider essentialhub proteins in PPI networks, and essential non-hub proteins are ignored. What’s more,the existing methods are based on the topological structure of PPI networks, so theprediction results are sensitive to noises in PPI networks and have small overlap. Topalliate this deficiency, we propose a new centrality measure ICFC to predict theessential non-hub proteins in PPI networks. The experimental results show that ICFChas higher prediction accuracy rate than the nine previous approaches. The number ofessential non-hub proteins identified by ICFC is1.22~3.75times as other methods.(6) There are a large amount of non-essential non-hub proteins in PPI networks.Why these non-essential non-hub proteins have no lethality and what’s the meaning oftheir existence to the whole cell and organism are unknown yet. To address this problem,we compare the differences between non-essential non-hub proteins and essential non-hub proteins in terms of the topology and evolution. The experimental results showthat non-essential non-hub proteins have lower centrality measures values, lowerclustering coefficient and higher evolutionary rate. The results prove that a lack ofnon-essential non-hub proteins will not affect PPI networks and non-essential non-hubproteins are more likely to evolve. These are reasons for unessentiality of non-essentialnon-hub proteins. In PPI networks, non-essential non-hub proteins don’t carry importantbiological functions and they maintain the stability of organism states and basicbiological processes by their redundancy.(7)We compare and analyse hub proteins and non-hub proteins systematically intopology, evolution and function. The results deepen our understanding of hub proteinsand non-hub proteins, and provide a general view of the relationship betweentopological structure and functional lethality in PPI networks.In summary, we explore the relationship between proteins’ topological structureand their functions in PPI networks, and find the particular topological, evolutionaryand functional characteristics of hub proteins and non-hub proteins, which caneffectively explain proteins’ lethality. We propose four new measures: n-ipc, n-neighbor,n-iep and n-f, which are based on proteins’ biological function and are not sensitive tonoises in the datasets. Besides, we present a new centrality measure ICFC to predictessential non-hub proteins. All these contributions should be helpful to advance researchon systems biology.