 Research article
 Open access
 Published:
Comprehensive hostpathogen proteinprotein interaction network analysis
BMC Bioinformatics volumeÂ 21, ArticleÂ number:Â 400 (2020)
Abstract
Background
Infectious diseases are a cruel assassin with millions of victims around the world each year. Understanding infectious mechanism of viruses is indispensable for their inhibition. One of the best ways of unveiling this mechanism is to investigate the hostpathogen proteinprotein interaction network. In this paper we try to disclose many properties of this network. We focus on human as host and integrate experimentally 32,859 interaction between human proteins and virus proteins from several databases. We investigate different properties of human proteins targeted by virus proteins and find that most of them have a considerable high centrality scores in human intra proteinprotein interaction network. Investigating human proteins network properties which are targeted by different virus proteins can help us to design multipurpose drugs.
Results
As hostpathogen proteinprotein interaction network is a bipartite network and centrality measures for this type of networks are scarce, we proposed seven new centrality measures for analyzing bipartite networks. Applying them to different virus strains reveals unrandomness of attack strategies of virus proteins which could help us in drug design hence elevating the quality of life. They could also be used in detecting host essential proteins. Essential proteins are those whose functions are critical for survival of its host. One of the proposed centralities named diversity of predators, outperforms the other existing centralities in terms of detecting essential proteins and could be used as an optimal essential proteinsâ€™ marker.
Conclusions
Different centralities were applied to analyze human proteinprotein interaction network and to detect characteristics of human proteins targeted by virus proteins. Moreover, seven new centralities were proposed to analyze hostpathogen proteinprotein interaction network and to detect pathogensâ€™ favorite host protein victims. Comparing different centralities in detecting essential proteins reveals that diversity of predator (one of the proposed centralities) is the best essential protein marker.
Background
Killing millions of humans, infectious diseases are the most brutal enemies of the entire history. Billions of dollars are spent to reveal the way hosts are infected by pathogens and their presumptive victims. HostPathogen proteinprotein interactions can be the best clue for initiating infection which have been studied in different pathogens [1,2,3,4,5,6,7,8]. Exploring molecular functions, biological processes, cellular compartment, common pathways and the other properties of host proteins targeted by pathogens can help us in infectious disease inhibition. Investigating common targets of different pathogens could help us to design multipurpose drugs.
Moreover, investigating proteinprotein interactions between human proteins (HPPIs) could help us to find virusesâ€™ potential HPs victims. To do that, HPPI network is analyzed by different centrality measures. In network analysis, centrality is the main concept of identifying gravity of each node in the network. Centrality measures can be used to find most important HPs in HPPI network (HPPIN) to identify new drug targets [9,10,11,12,13,14,15]. Each centrality measure defines nodesâ€™ weight from a different perspective. HPPIN has been analyzed by different centralities such as Degree Centrality [16], Closeness [17], Lobby Index [18], Betweenness [19], Clustering Coefficient [20], Leader Rank [21], Topological Coefficient [22], Module Centrality [23], Eigenvector Centrality [24], Neighborhood Connectivity [25], Normalized Alpha Centrality [26], Average Shortest Path Length [27], Subgraph Centrality [28], Radiality [29], Range limited Centrality [30] and Eccentricity [31].
Essential genes are minimal gene sets which are indispensable for a living cell and their functions are the foundation of life [32]. As disruption of these genes can lead to cell death, we investigate human virus proteinprotein interaction network (HVPPIN) to see whether the product of essential genes (essential proteins) are main targets of virus proteins (VPS) and which human proteins (HPs) are targeted by more VPs.
In this paper, we focus on human as host and integrate experimentally proteinprotein interactions (PPIs) between human proteins and virus proteins (HVPPIs) from different databases. Exploring the HVPPIN shows that human proteins which are targeted by different virus strains either have a lot of interactors in intra HPPIN or bridge between two large cliques.
Additionally, network analysis is performed on HPPIN by eight different centralities. Results demonstrate that centrality scores of HPs targeted by different virus strains are significantly higher than the other HPs. Besides, it reveals that centrality scores of essential proteins (EPs) are significantly higher than nonessential proteins.
HVPPIN is a special type of network called bipartite in which interactions are interspecies in contrast with HPPIN which is a unipartite network meaning interactions are intraspecies. As most of the common centralities are designed for unipartite network and HVPPIN is a bipartite network, seven novel bipartite centralities including CHTV (connectivity of human proteins targeted by same virus protein), PS (propagation speed), DP (diversity of predators), DSP (decreased shortest path), CI (component index), CR (crown centrality) and VC (vulnerable centrality) are proposed to analyze HVPPIN.
CHTV and CR scores of HVPPIN demonstrate the significant higher scores of real samples in comparison with random ones. PS analysis reveals the property of three degree of separation for HPPIN in presence of virus proteins. DSP results disclose the robustness of HPPIN. By doing extra analysis on VC, we found that virus proteins that have higher VC score, target considerably more EPs in comparison with virus proteins with lower VC score and could be chosen for drug target purposes. Finally, comparing DP scores of EPs and nonEPs with other centrality scores of EPs and nonEPs reveals that DP out performs the other centralities in detecting EPs and could be used as EPs marker.
Moreover, we make a database for human proteins centralities, DP score and virus families evolutionary distance which is publicly available at http://bioinf.modares.ac.ir/software/PHINA.
Methods
Preliminaries and definitions
A graph is generally illustrated by G(V,â€‰E) where V(G) is the set of nodes also known as vertex set and E(G)â€‰âŠ‚â€‰V(G)â€‰â¨¯â€‰V(G) is the set of edges between the graphâ€™s nodes also known as edge set. Two vertices v_{i} and v_{j} which are connected by an edge are called adjacent. All the vertices adjacent to v_{i} are called the neighbors of v_{i} and declared by N(v_{i}). Number of neighbors of v_{i} is called degree of v_{i} and denoted by deg(v_{i})â€‰=â€‰N(v_{i})âˆ£. Edges which have the same end vertices are called parallel edges and if both end vertices are the same, it will be called a selfloop. A graph which has neither parallel edges nor a selfloop is called a simple graph, otherwise it is called multigraph or hyper graph. A sequence of alternating vertices is called walk. Walk with no repeated edge is called trial and trial with sets of unique vertices is called path. If a path exists between each pair of graph vertices, it is called connected graph. Subgraph of a graph is a graph S(V,â€‰E) which V(S)â€‰âŠ†â€‰V(G) and E(S)â€‰âŠ†â€‰E(G). Maximally connected subgraph of a graph is called component. A fully connected graph is said complete graph. A complete subgraph of a graph is called clique [33].
A bipartite graph [34] G(T,â€‰B,â€‰E) is a kind of graph that has two vertices disjointing subset T and B (top and bottom) which V(G)â€‰=â€‰T(G)â€‰â‹ƒâ€‰B(G) and T(G)â€‰â‹‚â€‰B(G)â€‰=â€‰âŒ€. Moreover, each edge has an end vertex from the top node and another from bottom nodes (âˆ€(tb)â€‰âˆŠâ€‰E(G)â€‰âˆ£â€‰tâ€‰âˆŠâ€‰T(G), b âˆŠB(G)). In a bipartite graph for any top vertex t_{i}â€‰âˆŠâ€‰T(G), N(t_{i})â€‰âŠ‚â€‰B(G) while for any bottom vertex b_{i}â€‰âˆŠâ€‰B(G), N(b_{i})â€‰âŠ‚â€‰T(G). There are many real bipartite networks such as actormovie network [35] which illustrates the movies and the actors playing in them, authorarticle network [36] which shows the articles and theirs authors, genedisease network [37,38,39,40,41,42,43] which reveals diseases and their targeted genes, metaboliteenzyme network [44,45,46] which shows metabolites and their corresponding enzymes, and hostpathogen PPI network such as fungal pathogen scedosporium aurantiacum with human long epithelial cells [47], Leptospira interrogans and Homo sapiens [48], extracellular bacterial pathogen and human host [49], zika virus nonstructural proteins and human host proteins [50], bacterial and fungal pathogens and maize stalks host [51].
Bipartite projection of graph G(T,â€‰B,â€‰E) is graph Gâ€‰â€²â€‰(B^{â€²},â€‰Eâ€²) where nodes are subsets of G bottom nodes which have at least one common neighbor. Each pair of nodes that have common neighbors are connected with an edge as it is shown in sample graph of Fig 1. In other words, each top node makes a clique with the size of its degree.
Human proteinprotein interaction network (HPPIN) is a graph, the nodes of which are human proteins (HPs) and interaction between them are illustrated by graph edges. There are many public repositories for molecular interaction data [52]. We extracted 262,814 HPPIs from Intact [53] and BioGrid [54] including 19,995 HPs. Most of biological networks have one main component. As it is shown in Fig. 2A, more than 0.999 of HPs are in one main component which reveals the existence of a path between 99.9% HP pairs in HPPIN.
Humanvirus proteinprotein interaction network (HVPPIN) is a bipartite graph in which virus and human proteins are its top and bottom nodes respectively and interaction between them are illustrated by graph edges. We extracted 34,768 HVPPIs from Intact [53], VirusMint [55], DIP [56], STRING [57], and BioGrid [54] including 7141 HPs and 1281 VPs which belong to 34 different families, 88 different types of genus and 236 different strains as depicted in Fig. 2B.
In the proposed article, we investigated the following famous centralities in HPPIN:
DC (Degree Centrality): For each HP, DC is the number of its interacting partners. Virus proteins can infect many of HPs by targeting an HP with a high degree.
NC (Neighborhood Connectivity): For each HP, NC is the average degree of all its neighbors. Virus proteins can infect many of HPs by targeting an HP with the high NC through its neighbors.
SP (Average Shortest Path Length): The length of a path is the number of interactions which will be traversed. The pass with minimum length between each two HPs is considered as the shortest path. For each HP as shown in the following formula, SP is the summation of the shortest path between that HP and all the other HPs divided by the number of HPs. \( {SP}_p=\left({\sum}_{m=1}^n{S}_{p,m}\right)/n \). HPs with low SP can be considered as good targets for virus proteins. By reaching those HPs, virus proteins can quickly propagate to the other HPs.
TC (Topological Coefficients): The extent to which an HP shares neighbors with others. Zero would be assigned to HPs that have less than two neighbors. TC is defined as: TC_{p}â€‰=â€‰avg(J_{p, m})/k_{p}, where J_{p. m} is defined for all HPs m that share at least one neighbor with HP p and the value J_{p. m} is the number of neighbors shared between the HPs p and m, plus one if there is a direct link between them. However, k_{p} is the degree of p.
CS (Closeness Centrality): The reciprocal of the average shortest path length. It is, in fact, a number between 0 and 1 which is computed as: CS_{p}â€‰=â€‰1/avg(L_{p, m}).
where L_{p. m} is the length of the shortest path between two HPs p and m.
Zero would be assigned to isolated HPs. This measure shows Propagation speed of viruses from a given HP to the other reachable ones in the HPPIN.
CC (Clustering Coefficients): For each HP, CC is the number of triangles passing through it, relative to the maximum number of triangles that could pass through. CC_{p}â€‰=â€‰2e_{p}/k_{p}(k_{p}â€‰âˆ’â€‰1)
where k_{p} is the degree of p and e_{p} is the number of connected pairs between all its neighbors.
BC (Betweenness Centrality): The amount of control which one HP exerts over the interactions of others in the HPPIN and it is defined as follows: BC_{p}â€‰=â€‰â€‰âˆ‘â€‰Ïƒ_{st}(p)/Ïƒ_{st}, where s and t are HPs in the HPPIN different from p, Ïƒ_{st} shows the number of shortest paths from s to t, and Ïƒ_{st} (p) is the number of shortest paths from s to t that p lies on.
RD (Radiality): For each HP, RD is calculated by subtracting SP from the HPPIN diameter. Hence, HPs with higher RD are usually closer to the other nodes, whereas, HPs with lower RD are peripheral.
EV (Eigenvector centrality): EV is calculated by the eigenvector of the largest eigenvalue of adjacency matrix. It is a measure to declare the influence level of an HP node within HPPIN. HPs with high EV have a widereaching influence on HPPIN.
PR (Page rank centrality): For each HP, it measures the importance of HPs connected to that HP. It is equal to the sum of page rank score of its neighbors.
As it is shown in Fig. 4C, for all of the 19,995 HPs of HPPIN, all of the mentioned centralities are calculated and reported in our database which is publicly accessible at the following address: http://bioinf.modares.ac.ir/software/PHINA/Centrality.php
Moreover, we propose seven new centralities for analyzing HVPPIN:
Connectivity of the human nodes targeted by the same virus node (CHTV)
For each virus node, human nodes targeted by that node (HTV) are chosen. CHTV is the number of connected pairs of HTVs in HPPIN over all possible pairs and reached by the following formula:
CHTV_{i} is connectivity of the human proteins targeted by ith virus protein, n is the number of HTV_{i}s and A_{jk} is a binary function with one value in the cases where an edge exists between H_{j} and H_{k} and zero otherwise.
As it is shown in Fig. 3A, all the targeted nodes by the same source node will construct a clique in projection network. Hence, CHTV will show the fraction of clique edges which contribute in real world network of targeted nodes.
Propagation speed
Propagation speed (PS) is a measure which illustrates the speed of propagation of a virus in HPPIN or a rumor in society. We defined three levels of PS. Each level indicates the percentage of infected human proteins in that round. So PS1 is the number of human proteins targeted by virus protein and their neighbors over the number of all human proteins in HPPIN. PS2 is the number of infected human proteins of level one plus all of their neighbors over the number of all human proteins in HPPIN. Finally, PS3 is the number of infected human proteins of level two plus all of their neighbors over the number of all human proteins in HPPIN.
As an example, consider node S as the first infected node in two different depicted networks of Fig. 3B. PS1, PS2 and PS3 for the first network (network at the top of the figure) is 0.3, 0.7 and 1, respectively. PS1, PS2 and PS3 for the second network (network at the bottom of the figure) is 0.2, 0.4 and 0.6, respectively. Score 1 for PS3 of the first network reveals that all of the network nodes are infected while score 0.6 shows that 60% of the second network nodes are infected.
In HPPIN, HTVs are chosen. Cumulative percentage of the number of HTVsâ€™ first neighbors, second neighbors and third neighbors are PS1, PS2 and PS3 scores, respectively. In other words, PS1, PS2 and PS3 indicate the percentage of human proteins of HPPIN which will be infected within the first, second and third round of interactions, respectively.
Among all 34,768 interactions in HVPPIN, 28639 interactions containing virus strains with at least 5 interactions are chosen. The final HVPPIN contains 1093 virus proteins belonging to 26 different families, 67 different types of genus and 124 different strains. For each strain separately, HTVs are chosen and PS1, PS2 and PS3 scores are calculated.
Diversity of predators
In HVPPIN, HPs, which are targets of VPs of different families, are chosen. For each of these HPs, each of the virus families is called a predator. Evolutionary distance (ED) among any couple of predators of each HP is calculated. For each HP, mean of EDs of its predators multiplied by the number of its predators is considered as its diversity of predators (DP) score. As an example, Q9Y6H1 is targeted by VPs of four different virus families (Matonaviridae (M), Flaviviridae (F), Herpesviridae (H) and Papillomaviridae (P)). To calculate its DP, we need to calculate \( \left(\begin{array}{c}n\\ {}2\end{array}\right) \) EDs between each couple of its predators where n is the number of predators. Therefore, for the mentioned example, we need to calculate \( \left(\begin{array}{c}4\\ {}2\end{array}\right)=6 \) Eds and the DP will be calculated as follows:
As there is not any database for ED among virus families, we create a database for calculating ED of the most famous virus families. To calculate ED between two virus families, we make NeedlemanWunsch [58] global pairwise alignment between each VP of the first family with all VPs of the second one. Mean of alignment scores of all pairs is considered as ED between the two families. As an example, for calculating ED between Orthomyxoviridae with 132 different VPs and Papillomaviridae with 106 VPs, 13,992 (132*106) global pairwise alignments were calculated and mean of all these alignments were reported as ED of these two families. Our database (Fig. 4B) is publicly accessible at the following address: http://bioinf.modares.ac.ir/software/PHINA/EvolutionaryDistance.php
Among all 6500 HPs targeted by different VPs in HVPPIN, 3141 HPs interact with VPs of at least two virus families. For all of these HPs, DP is calculated and reported in our database (Fig. 4A) which is publicly accessible at the following address: http://bioinf.modares.ac.ir/software/PHINA/VirusFamilies.php
(All the pictures of virus families inside our site are gathered from ViralZone:www.expasy.org/viralzone, SIB Swiss Institute of Bioinformatics [59] which is licensed under a creative common attribution 4.0 international.)
Decreased shortest path
In the main HPPIN component the shortest path between each two HP pair is calculated. Afterwards, HVPPIs of one of the virus strains is added to HPPIN and the shortest path between each two HP pairs is recalculated. Difference between sum of all possible HP pairsâ€™ shortest path in presence or absence of HVPPIs is measured as decreased shortest path (DSP).
Among all 32,859 interactions in HVPPIN, 28306 interactions containing virus proteins with at least 2 interactions are chosen. The final HVPPIN contains 784 virus proteins belonging to 25 different families, 64 different types of genus and 111 different strains. For each of these strains, DSP is calculated and DSPsâ€™ mean of each familyâ€™s strain is considered as DSP of that family. The same was performed for random cases in a way that random nodes with the same degree distribution of real strains were added to HPPIN and DSP were calculated.
Component index
Component index is the tendency of HPs targeted by VPs of one genus to interact with each other rather than the HPs targeted by VPs of another genus. First, induced sub graph (ISG) of HPs targeted by all VPs of each strain was extracted from HPPIN. Then, for each strain, the number of interactions within its ISG HPs, and the number of interactions between its HPs and other ISG HPs was calculated. Finally, component index was defined between two strains by the following formula:
n_{ic} is the number of first ISG HPs, n_{oc} is the number of second ISG HPs, e_{ij} is 1 if ith HP of first ISG interacts with jth HP of the first ISG. e_{io} is 1 if ith HP of first ISG interacts with oth HP of second ISG. In other words, component index is the difference between the intrainteractions of an ISG and interinteractions of that ISG with another ISG over the sum of inter and intra interactions.
Crown centrality
Virus proteins which target the same pair of HPs formed a crown. Number of virus proteins which target the same pair of HPs determine the number of ivories of a crown (Fig. 5A). Number of ivories of a crown equals the degree of node reached by vertex contraction of HP pairs. Existence of crowns in HVPPIN shows the tendency of virus proteins to be cooperative in attacking HPs. Crowns with higher number of ivories shows the more important interaction between HP pairs of that crown. HP pairs of a crown which are adjacent in HPPIN has the clustering coefficient 1.
Among all 34,768 interactions in HVPPIN, 27859 interactions containing virus proteins with a degree of more than one are chosen. The final HVPPIN contains 990 virus proteins belonging to 19 different families, 42 different types of genus and 67 different strains.
We investigated the final HVPPIN for the crowns and did the same for random models with the same degree distribution.
Vulnerable centrality
As projection of a bipartite network is a simple graph and there are many different available centralities to analyze it, one way of analyzing bipartite network is analyzing its projection.
The main problem of this idea is the information loss in converting a bipartite network to its projection.
We define vulnerable centrality to reflect the effect of each top node in bottomprojection. As an example, as it is shown in Fig. 5B, edge T_{1}T_{4} is just made by S_{2} while T_{1}T_{3} is made by both S_{1} and S_{3} in projection. Thus, by elimination of S_{3}, edge T_{1}T_{3} remains, though by elimination of S_{2}, edge T_{1}T_{4} will be permanently dropped from projection.
Vulnerable centrality is defined for each top node as sum of reciprocal of projection weight divided by the equivalent clique size. As an example, for the graph of Fig. 5B, V_{1}. V_{2}. V_{3} (vulnerable score for S_{1}. S_{2}. S_{3}) is calculated by:
Essential proteins functions are indispensable for life cycle and thus, investigating them can be useful for both drug target purposes for inhibition and better understanding of their behaviors. 2472 EPs were gathered from Georgi [60], 1734 EPs were reported in Blomen [61], 1878 EPs were extracted from Wang [62], 3230 EPs were derived from Lek [63] and 7127 were collected from Chen [64]. Finally, 6115 genes which were at least reported in two of the mentioned studies were considered as essential proteins.
Lots of papers apply network topology for detecting new essential proteins [65,66,67,68,69,70]. Eight different mentioned centralities and two new proposed centralities (DP and VC) were calculated for all of the HPs of HPPIN. Afterwards, centralities of essential HPs were compared against nonessential HPs and the results were reported to investigate whether they could be applied for detecting new essential proteins.
Furthermore, Virus proteins of each strain, were clustered into 2 groups based on vulnerable score. Thereafter, the number of essential proteins targeted by each group were compared to see whether there exists a significant difference between the two groups.
Results
All the existing mentioned centralities and the two new proposed centralities (DP and VC) are investigated to see whether they could be an essential protein marker.
For the other five proposed centralities, different topics were investigated. For CHTV and crown centrality, we figured out if there is a significant difference in real sample scores in comparison with random sample scores or not. We also uncovered the degree of separation of HPPIN. Moreover, robustness of HPPIN is measured by DSP. Propensity of HPs targeted by the same VP to have inter or intra interaction was investigated.
Essential proteins are critical nodes in HPPIN
Ten different centralities were measured in HPPIN and as it is shown in Table 1 and Fig. 6, essential proteinsâ€™ centrality scores were significantly different in comparison with nonessential proteins.
Virus proteins with higher VC scores have a huge tendency to target essential proteins
In our model, vulnerable centrality illustrates the effect of each virus protein in constructing the projection of infected human proteins. Thus, virus proteins with higher vulnerable scores must be inhibited sooner than the others. In other words, inhibiting the virus proteins with higher vulnerable scores will maximized the chance of early inhibition of whole virus spread. As it is illustrated in Fig. 7, by calculating vulnerable scores of 29 different strains, virus proteins of each strain have a considerable diverse vulnerable score which makes this centrality capable of detecting the important virus proteins of each strain.
Moreover, virus proteins of each strain were separated into two groups according to VC scores. Essential HPs targeted by each group are reported in Table 2.
As clearly manifested by the results, virus proteins with higher VC scores have a huge tendency to target essential proteins. Consequently, VPs with higher VC scores seem to be better drug targets in comparison with others VPs.
DP centrality outperformed the other centralities in detecting essential proteins and potential drug targets
Among all 6500 HPs targeted by different VPs in HVPPIN, 3141 HPs interact with VPs of 34 different virus families. Table 3 shows 15 HPs with the highest DP score.
Further analysis reveals that EPsâ€™ DP scores are significantly higher than nonEPsâ€™ DP score and it could be used as a marker for detecting EPs.
To prove our claim, we compare the ability of detecting EPs between the proposed centrality and all the other mentioned centralities. To do so, for each centrality, we picked 200 HPs with the highest score and counted the number of EPs to declare the sensitivity of that centrality. As it is shown in Table 4, DP centrality outperformed others in detecting EPs.
To check the ability of different centralities in detecting potential drug targets, we chose two different gene expression data (GSE1739, GSE150316). Two hundred ninety nine and one hundred sixty two differentially expressed genes (DEG) were extracted from GSE150316 and GSE1739 respectively. Different centrality measures were calculated for them. In each centrality, the number of DEGs with centrality scores more than 3rd quantile score were reported in Table 5. This reveals the ability of each centrality in detecting potential drug targets.
CHTV scores of real samples are considerably higher than random cases
By considering n as the number of different viruses, n CHTV scores have been separately calculated for each genus. Mean of CHTV score of each type has been reported as CHTV score of that type.
Moreover, 64 Random classes were constructed equal to the number of genus types. For each class, CHTV_{i} shown in Fig. 8, is calculated for each virus protein of the related class by selecting random human proteins from human proteins of HPPIN equal to the number of HTV_{i}s.
Furthermore, for comparing the results of CHTV scores of real and random samples, statistics score (zscore) is calculated for each class of virus genus by the following formula:
where Z_{i} and CHTV_{i} indicate the zscore and CHTV score of ith genus, respectively and CHTVâ€‰âˆ’â€‰Random_{i} shows the CHTV score of ith related genus.
The standard score Z_{i} shows the distance of CHTV score of ith genus from the mean of CHTV score of the related random class. As it is shown in Fig. 9, 83% of zscores are more than 3 times of standard deviation above mean of random CHTV scores.
Number of crowns with more than two ivories of real samples are substantive in comparison with random classes
Results are depicted in Fig. 9 for crowns with one, two, three and more than three ivories. As evident in Fig. 10, real model precedes random model in terms of percentage of crowns as well as the higher the number of ivories and the greater difference between real and random model.
More than 70% of HPs are infected in just two rounds (two degree of separation)
For each HVPPIN strain separately, HTVs are chosen and PS1, PS2 and PS3 scores are calculated. For each family, mean of each score is illustrated in Fig. 11 with red color.
Moreover, 26 classes containing 124 random classes were constructed equal to the number of families and strains. For each class, equal to the number of virus proteinsâ€™ targets of the related strain, random human proteins were selected as HTVs and PS1, PS2 and PS3 scores were calculated. For each family, mean of each score was illustrated in Fig. 10 with teal color.
Only in two rounds, in most of the virus families, more than 70% of HPs of HPPIN are infected. Furthermore, for comparing the results of PS scores of real and random samples, statistics score (zscore) was calculated for each class of virus as shown in Fig. 12.
DSP illustrates the robustness of HPPIN
WattsStrogats model [20] is a random model constructed by rewiring the edges of a regular ring lattice. Shortest path between pairs of a regular ring lattice is so larger than WattsStrogats model because rewired edges act as shortcuts. As shown in Fig. 13, random classes thus far, have better DSP scores which is due to the shortcuts created by random choosing of HPs in random model, while in real models, as virus proteins target specific HPs in specific tissue, local shortest path reduction will happen. This result also shows the robustness of HPPIN.
Most of strains have a positive component index
Component Index (CI) was calculated between all 1431 pairs of 54 strains which had at least 5 HVPPIs. Mean and median of all CIs were 0.62 and 0.71, respectively, which shows the great tendency of HPs targeted by each strain to interacts within its ISG and make most of them as a real component. As an example, in Table 6, Alphainfluenzavirus CI is calculated with the first 10 largest genera (in terms of the number of interactions). Alphainfluenzavirus has 2794 HPs and 27,296 interactions. The number of HPs and interactions of the other strains are placed in OCHPs and OCIs columns, respectively.
Discussion
Figure 14 summarizes the selections of the most central host proteins by different measures. For each of the 11 centrality measures worked in this article, 200 HPs with the highest centrality scores were selected. Degree, closeness, average shortest path, and page rank centralities had the most common HP targets among their first 200 highest score HPs.
By investigating all the centrality measures, we found that all of the top 10 HPs with the highest centrality scores in different centrality measures are essential proteins.
Moreover, the following HPs were detected as HPs with high centrality score in the most of the centrality measures:

P05412 (Transcription factor AP1)

P62993 (Growth factor receptorbound protein 2)

P08238 (Heat shock protein HSP 90beta)

Q99459 (Cell division cycle 5like protein)

Q08379 (Golgin subfamily A member 2)

P61981 (14â€“33 protein gamma)

Q86VP6 (Cullinassociated NEDD8dissociated protein 1)

P10809 (60â€‰kDa heat shock protein, mitochondrial)

P11142 (Heat shock cognate 71â€‰kDa protein)

P27824 (Calnexin)
Connectivity of HPs targeted by the same VP and propagation speed of viruses in HPPIN reveals that VPs select their targets purposefully.
The tendency of virus proteins with high VC scores in targeting essential proteins and the results of DP centrality in detecting essential proteins shows that HPs with high DP score can be considered as potential drug targets.
Investigating Crown centrality scores reveals the tendency of VPs to have collaboration in targeting same HP pairs.
Finally, Table 7 summarizes some of the possible usages of the proposed centralities.
Conclusions
In this article, we studied the properties of a bipartite network generated from interactions between human proteins versus virus proteins. As there are different virus families, we investigated each family as a separate network and reported the results for most famous virus families. As HVPPIN is a bipartite network and centrality measures for this type of network is scarce, seven new centralities were proposed on HVPPIN and measured on different strains of famous virus families. In all proposed centralities, significant difference was observed between real and random samplesâ€™ scores.
Moreover, we found some significant properties of essential proteins. By investigating HPPIN and calculating ten famous centralities, it was revealed that essential HPs have a considerable higher centrality scores in comparison with nonessential HPs and it could be used for finding new essential HPs. In addition, we observed that DP scores have the same pattern. For finding the best marker of EPs, for each of the centralities, we select 200 HPs with the highest scores and calculated the sensitivity of detecting EPs with them. Results demonstrate that DP outperforms the others. Furthermore, analyzing VC scores of HVPPIN disclosed that VPs with high VC scores target essential HPs significantly higher than VCs with low VC scores. This observation could be used for choosing VCs with high VC score as the first target of drug design.
The current work can be extended by using the proposed centralities in the other bipartite networks. Moreover, it is suggested to recalculate centralities by adding new HPPIs to HPPIN for finding potential EPs. Doing the same with HVPPIN and considering new HP targets of highVCscores VPs and HPs with high DP scores as potential EPs are also recommended.
Availability of data and materials
Some part of data, analyzed during the current study are publicly available in the http://bioinf.modares.ac.ir/software/PHINA
The other datasets which were analyzed during the current study available from the corresponding author on reasonable request.
Abbreviations
 HPPI:

Human proteinprotein interaction
 HVPPI:

Humanvirus proteinprotein interaction
 HPPIN:

Human proteinprotein interaction network
 HVPPIN:

Humanvirus proteinprotein interaction network
 HP:

Human protein
 VP:

Virus protein
 EP:

Essential protein
 CHTV:

Connectivity of human proteins targeted by same virus protein
 PS:

Propagation speed
 DP:

Diversity of predators
 DSP:

Decreased shortest path
 CI:

Component index
 CR:

Crown centrality
 VC:

Vulnerable centrality
 DC:

Degree centrality
 NC:

Neighborhood connectivity
 SP:

Average shortest path length
 TC:

Topological coefficients
 CS:

Closeness centrality
 CC:

Clustering coefficients
 BC:

Betweenness centrality
 RD:

Radiality
 EV:

Eigenvector centrality
 PR:

Page rank centrality
 HTV:

Human proteins targeted by virus
 ED:

Evolutionary distance
 ISG:

Induced sub graph
 DEG:

Differentially expressed gene
References
Davis FP, Barkan DT, Eswar N, McKerrow JH, Sali A. Hostâ€“pathogen protein interactions predicted by comparative modeling. Protein Sci. 2007;16(12):2585â€“96.
Dyer MD, Murali TM, Sobral BW. Computational prediction of hostpathogen proteinâ€“protein interactions. Bioinformatics. 2007;23(13):i159â€“66.
Eng CLP, Tong JC, Tan TW. Predicting host tropism of influenza a virus proteins using random forest. BMC Med Genet. 2014;7(3):S1.
Evans P, Dampier W, Ungar L, Tozeren A. Prediction of HIV1 virushost protein interactions using virus and host sequence motifs. BMC Med Genet. 2009;2(1):27.
Barnes B, et al. Predicting novel proteinprotein interactions between the HIV1 virus and Homo sapiens. In: Student conference (ISC), 2016 IEEE EMBS international; 2016. p. 1â€“4.
Hale BG, Jackson D, Chen YH, Lamb RA, Randall RE. Influenza a virus NS1 protein binds p85Î² and activates phosphatidylinositol3kinase signaling. Proc Natl Acad Sci. 2006;103(38):14194â€“9.
Cui G, Fang C, Han K. Prediction of proteinprotein interactions between viruses and human by an SVM model. BMC Bioinformatics. 2012;13(7):S5.
Khorsand B, Savadi A, Zahiri J, Naghibzadeh M. Alpha influenza virus infiltration prediction using virushuman proteinâ€“protein interaction network. Math Biosci Eng. 2020;17(4):3109â€“29. https://doi.org/10.3934/mbe.2020176.
Miryala SK, Ramaiah S. Exploring the multidrug resistance in Escherichia coli O157: H7 by gene interaction network: a systems biology approach. Genomics. 2019;111(4):958â€“65.
Miryala SK, Anbarasu A, Ramaiah S. Impact of bedaquiline and capreomycin on the gene expression patterns of multidrugresistant mycobacterium tuberculosis H37Rv strain and understanding the molecular mechanism of antibiotic resistance. J Cell Biochem. 2019;120(9):14499â€“509.
Miryala SK, Anbarasu A, Ramaiah S. Systems biology studies in Pseudomonas aeruginosa PA01 to understand their role in biofilm formation and multidrug efflux pumps. Microb Pathog. 2019;136:103668.
Miryala SK, Anbarasu A, Ramaiah S. Role of SHV11, a class a Î²lactamase, gene in multidrug resistance among Klebsiella pneumoniae strains and understanding its mechanism by gene network analysis. Microb Drug Resist. 2020;26:900â€“8.
Miryala SK, Anbarasu A, Ramaiah S. Evolutionary relationship of penicillinbinding protein 2 coding penA gene and Understanding the role in drugresistance mechanism using gene interaction network analysis. In: Emerging Technologies for Agriculture and Environment. Singapore: Springer; 2020. p. 9â€“25.
Debroy R, Miryala SK, Naha A, Anbarasu A, Ramaiah S. Gene interaction network studies to decipher the multidrug resistance mechanism in salmonella enterica serovar Typhi CT18 reveal potential drug targets. Microb Pathog. 2020;142:104096.
Naha A, Miryala SK, Debroy R, Ramaiah S, Anbarasu A. Elucidating the multidrug resistance mechanism of Enterococcus faecalis V583: a gene interaction network analysis. Gene. 2020. p. 144704â€“16.
Proctor CH, Loomis CP. Analysis of sociometric data. Res Methods Soc relations. 1951;2:561â€“85.
Newman MEJ. A measure of betweenness centrality based on random walks. Soc Networks. 2005;27(1):39â€“54.
Korn A, Schubert A, Telcs A. Lobby index in networks. Phys A Stat Mech its Appl. 2009;388(11):2221â€“6.
Freeman LC. A set of measures of centrality based on betweenness. Sociometry. 1977;40:35â€“41.
Watts DJ, Strogatz SH. Collective dynamics of â€˜smallworldâ€™ networks. 1998;393:440â€“2.
LÃ¼ L, Zhang YC, Yeung CH, Zhou T. Leaders in social networks, the delicious case. PLoS One. 2011;6(6):e21202.
Stelzl U, et al. A human proteinprotein interaction network: a resource for annotating the proteome. Cell. 2005;122(6):957â€“68.
KovÃ¡cs IA, Palotai R, Szalay MS, Csermely P. Community landscapes: an integrative approach to determine overlapping network module hierarchy, identify key nodes and predict network dynamics. PLoS One. 2010;5(9):e12528.
Bonacich P. Power and centrality: a family of measures. Am J Sociol. 1987;92(5):1170â€“82.
Maslov S, Sneppen K. Specificity and stability in topology of protein networks. Science (80 ). 2002;296(5569):910â€“3.
Gosh R, Lerman K. A parameterized centrality metrics for network analysis. Phys Ther Rev. 2011. p. 66118â€“27.
Beardwood J, Halton JH, Hammersley JM. The shortest path through many points. Math Proc Camb Philos Soc. 1959;55(4):299â€“327.
Estrada E, RodriguezVelazquez JA. Subgraph centrality in complex networks. Phys Rev E. 2005;71(5):56103.
Brandes U. A faster algorithm for betweenness centrality. J Math Sociol. 2001;25(2):163â€“77.
ErcseyRavasz M, Lichtenwalter RN, Chawla NV, Toroczkai Z. Rangelimited centrality measures in complex networks. Phys Rev E. 2012;85(6):66103.
Hage P, Harary F. Eccentricity and centrality in networks. Soc Networks. 1995;17(1):57â€“63.
Itaya M. An estimation of minimal genome size required for life. FEBS Lett. 1995;362(3):257â€“60.
Diestel R. Graduate texts in mathematics: graph theory, vol. 173. Heidelb: SpringerVerlag; 2000.
Bondy JA, Murty USR. Graph theory with applications, vol. 290. Ontario: Citeseer; 1976.
Guillaume JL, Latapy M. Bipartite structure of all complex networks. Inf Process Lett. 2004;90(Issue5):215â€“21.
Gustafsson H, Hancock DJ, CÃ´tÃ© J. Describing citation structures in sport burnout literature: a citation network analysis. Psychol Sport Exerc. 2014;15(6):620â€“6.
Abdelmoneim AH, Mustafa MI, Mahmoud TA, Murshed NS, Hassan MA. In silico analysis and modeling of novel pathogenic single nucleotide polymorphisms (SNPs) in human CD40LG gene. bioRxiv. 2019:552596.
Ã–zgÃ¼r A, Vu T, Erkan G, Radev DR. Identifying genedisease associations using centrality on a literature mined geneinteraction network. Bioinformatics. 2008;24(13):i277â€“85.
Yuan F, Zhang YH, Kong XY, Cai YD. Identification of candidate genes related to inflammatory bowel disease using minimum redundancy maximum relevance, incremental feature selection, and the shortestpath approach. Biomed Res Int. 2017, 2017.
Zickenrott S, Angarica VE, Upadhyaya BB, Del Sol A. Prediction of diseaseâ€“geneâ€“drug relationships following a differential network analysis. Cell Death Dis. 2016;7(1):e2040.
Zeng X, Ding N, RodrÃguezPatÃ³n A, Zou Q. Probabilitybased collaborative filtering model for predicting geneâ€“disease associations. BMC Med Genet. 2017;10(5):76.
Hwang S, et al. HumanNet v2: human gene networks for disease research. Nucleic Acids Res. 2018;47(D1):D573â€“80.
Huang L, Wang Y, Wang Y, Bai T. Genedisease interaction retrieval from multiple sources: a network based method. Biomed Res Int. 2016;2016.
NodaGarcia L, Liebermeister W, Tawfik DS. Metaboliteâ€“enzyme coevolution: from single enzymes to metabolic pathways and networks. Annu Rev Biochem. 2018;87:187â€“216.
Ravasz E, Somera AL, Mongru DA, Oltvai ZN, BarabÃ¡si AL. Hierarchical organization of modularity in metabolic networks. Science (80 ). 2002;297(5586):1551â€“5.
Diether M, Sauer U. Towards detecting regulatory proteinâ€“metabolite interactions. Curr Opin Microbiol. 2017;39:16â€“23.
Kaur J, et al. Interactions of an emerging fungal pathogen Scedosporium aurantiacum with human lung epithelial cells. Sci Rep. 2019;9(1):5035.
Kumar S, Lata KS, Sharma P, Bhairappanavar SB, Soni S, Das J. Inferring pathogenhost interactions between Leptospira interrogans and Homo sapiens using network theory. Sci Rep. 2019;9(1):1434.
Griesenauer B, et al. Determination of an interaction network between an extracellular bacterial pathogen and the human host. MBio. 2019;10(3):e01193â€“19.
Golubeva VA, et al. Network of interactions between ZIKA virus nonstructural proteins and human host proteins. Cells. 2020;9(1):153.
CoboDÃaz JF, Baroncelli R, Le Floch G, Picot A. Combined metabarcoding and cooccurrence network analysis to profile the bacterial, fungal and fusarium communities and their interactions in maize stalks. Front Microbiol. 2019;10:261.
Miryala SK, Anbarasu A, Ramaiah S. Discerning molecular interactions: a comprehensive review on biomolecular interaction databases and network analysis tools. Gene. 2018;642:84â€“94.
Kerrien S, et al. The IntAct molecular interaction database in 2012. Nucleic Acids Res. 2011;40(D1):D841â€“46.
Oughtred R, et al. The BioGRID interaction database: 2019 update. Nucleic Acids Res. 2018;47(D1):D529â€“41.
Chatraryamontri A, et al. VirusMINT: a viral protein interaction database. Nucleic Acids Res. Jan. 2009;37(Database issue):D669â€“73.
Deane CM. Protein interactions: two methods for assessment of the reliability of high throughput observations. Mol Cell Proteomics. 2002;1(5):349â€“56.
Cook H, Doncheva N, Szklarczyk D, von Mering C, Jensen L. Viruses. STRING: a virushost proteinprotein interaction database. Viruses. 2018;10(10):519.
Needleman SB, Wunsch CD. A general method applicable to the search for similarities in the amino acid sequence of two proteins. J Mol Biol. 1970;48(3):443â€“53.
Masson P, et al. ViralZone: recent updates to the virus knowledge resource. Nucleic Acids Res. 2012;41(D1):D579â€“83.
Georgi B, Voight BF, BuÄ‡an M. From mouse to human: evolutionary genomics analysis of human Orthologs of essential genes. PLoS Genet. 2013;9(5):e1003484.
Blomen VA, et al. Gene essentiality and synthetic lethality in haploid human cells. Science (80 ). 2015;350(6264):1092â€“6.
Wang T, et al. Identification and characterization of essential genes in the human genome. Science (80 ). 2015;350(6264):1096â€“101.
Lek M, et al. Analysis of proteincoding genetic variation in 60,706 humans. Nature. Aug. 2016;536:285.
Chen WH, Lu G, Chen X, Zhao XM, Bork P. OGEE v2: an update of the online gene essentiality database with special focus on differentially essential genes in human cancer cell lines. Nucleic Acids Res. Oct. 2016;45(D1):D940â€“4.
Li X, Li W, Zeng M, Zheng R, Li M. Networkbased methods for predicting essential genes or proteins: a survey. Brief Bioinform. 2020;21(2):566â€“83.
Li G, Li M, Wang J, Wu J, Wu FX, Pan Y. Predicting essential proteins based on subcellular localization, orthology and PPI networks. BMC Bioinformatics. 2016;17(8):279.
Jalili M, et al. Evolution of centrality measurements for the detection of essential proteins in biological networks. Front Physiol. 2016;7:375.
Li M, Ni P, Chen X, Wang J, Wu F, Pan Y. "Construction of Refined Protein Interaction Network for Predicting Essential Proteins," in IEEE/ACM Transactions on Computational Biology and Bioinformatics. 2019;16(4):1386â€“97. https://doi.org/10.1109/TCBB.2017.2665482.
Qin C, Sun Y, Dong Y. A new method for identifying essential proteins based on network topology properties and protein complexes. PLoS One. 2016;11(8):e0161042.
Azhagesan K, Ravindran B, Raman K. Networkbased features enable prediction of essential genes across diverse organisms. PLoS One. 2018;13(12):e0208722.
Acknowledgements
Not applicable
Funding
There is not any special source of funding
Author information
Authors and Affiliations
Contributions
BK. designed the work, constructed and analyzed the datasets, proposed new centralities. AS. helped in data acquisition and interpretation. MN. helped in conception of the work and interpretation of the data. The authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Ethics approval and consent to participate
Not applicable
Consent for publication
Not applicable
Competing interests
The authors declare that they have no competing interests.
Additional information
Publisherâ€™s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
About this article
Cite this article
Khorsand, B., Savadi, A. & Naghibzadeh, M. Comprehensive hostpathogen proteinprotein interaction network analysis. BMC Bioinformatics 21, 400 (2020). https://doi.org/10.1186/s1285902003706z
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s1285902003706z