A Graph-Based Approach for Finding the Dengue Infection Pathways in Humans Using Protein

Abstract

Dengue virus (DENV) is one of the deadly arboviruses, which is primarily transmitted by Aedes aegypti, and causes dengue infection to the humans. According to WHO, every year around 390 million humans are affected by DENV, of which around 50 million deaths are reported. Knowledge of the various diseases caused by the DENV would greatly encourage to understand the infection mechanism and help to design new antiviral drug discovery. We propose a quasi-clique and quasi-biclique algorithm to classify infection gateway proteins of the human body and possible pathways of DENV leading to various diseases. For this, we have examined three networks, dengue-human protein–protein interaction network, human protein interaction network, and human proteins-disease association network. The prediction result states that DENV may lead to various diseases in the human body, including cancer, asthma, ulcerative colitis, multiple sclerosis, premature birth, and so on. Some of the results have recently been validated experimentally. This study may endow with potential targets for more effective anti-dengue remedial contribution.

1. Introduction

Dengue virus (DENV) is a growing threat to worldwide human being health. It consists of four serotypes, DENV-1, DENV-2, DENV-3, and DENV-4. Each of these serotypes causes the same disease. But, some serotypes, for example, DENV2, are linked with more severe dengue disease. All these serotypes have different types of interactions with the antibodies in human blood. The viral genome codes for 10 proteins. Three are structural proteins, capsid (C), the precursor of membrane protein (PrM/M), and envelope protein (E), while the rest are nonstructural (NS) proteins—NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5.

The DENV has the capability to survive and replicate in both human and mosquito host organisms. Because the dengue genome encodes only 10 viral proteins, the virus needs to manipulate its hosts at a molecular level (Doolittle and Gomez, 2011). So, virus-host protein–protein interaction (PPI) enlightens the function of virus proteins and how they reproduce and cause diseases. Several studies are reported to show PPI between dengue proteins and human proteins. Moreover, DENV is capable of causing several other diseases in the human body. Some medical reports state that the patients affected by DENV result in other forms of the illness like organ damage, severe bleeding, dehydration, and even death. So, complete pathway analysis of DENV is important to understand its functionality in the human body.

Various researches are going on for predicting pathway analysis of different diseases. In Yamanishi et al. (2004), a variant of the kernel-based method is used for pathway protein interaction. Some databases have used Bayesian classification for PPI prediction in yeast (Chowdhary et al., 2009). But most of these techniques are applied within a single organism, such as yeast, human, and so on. However, identification of PPI among viral proteins and human proteins and pathway analysis of the interacting human protein will help to predict possible disease that may occur due to infection of particular viral protein.

In Khadka et al. (2011), yeast two-hybrid test is used to screen new interactions between DENV and human proteins. The article also demonstrates that DENV like other diseases preferentially interacts with the proteins that are centrally located in the human protein interaction network. It is also reported that the human proteins that interact with dengue proteins also interact with hepatitis C virus (HCV) and HIV virus proteins. But experimental methods are very time consuming and costly due to the requirement of expensive laboratory equipment. So, it reveals only a small part of the all possible disease occurred due to DENV infection. This leads to the application of computational methods to investigate the pathways of dengue disease.

The main objective of this article is to predict the diseases that may occur due to DENV infection in the human body. We have used a quasi-clique and quasi-biclique algorithm to find the gateway proteins of infection of the human interactome. Analysis of gateway proteins is very important, as they are targeted by the viral proteins for entering and affecting a particular cell. Three networks are considered for generating the gateway proteins. First, the dengue–human PPI, which stores interactions between dengue proteins and human proteins. Second, human PPI, which depicts the interactions between the human proteins, present in the first set, with the rest of the human proteins. Third, human-disease association network, which portrays the interaction between human proteins with different diseases. Then the quasi-biclique algorithm is applied to dengue–human PPI to get the strong interconnection network. The quasi-clique algorithm is applied to human PPI database to identify the human proteins that overlap with the human proteins of quasi-bicliques. These overlapping proteins are called gateway proteins. Using the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway and gene ontology (GO) analysis, the complete pathway of these gateway proteins is analyzed.

Furthermore, we applied the quasi-biclique algorithm on human-disease association network corresponding to each quasi-clique. The overlapping proteins are also treated as gateway proteins. Thus, two sets of gateway proteins are generated in this study. There are some diseases that are directly associated with human proteins, which interact with different dengue proteins. These direct interactions are also reported in this article.

2. Materials and Methods

2.1. Databases and preprocessing

In this section, the collection and preprocessing of the datasets have been described below.

2.1.1. Dengue–human protein interaction database: DenvInt

Dengue–human PPI database-DenvInt (https://denvint.000webhostapp.com) is composed of 545 nonredundant interactions between 341 different human proteins and 10 dengue proteins (Dey and Mukhopadhyay, 2017). The database is prepared by considering dengue–human PPI published in peer-reviewed journals and virus databases like VirHostNet and VirusMentha. Only experimentally validated interactions like yeast two-hybrid test (Folly et al., 2011; Khadka et al., 2011; Le Breton et al., 2011; Dumrong Mairiang et al., 2012), bacterial two-hybrid test (Folly et al., 2011) RNA affinity chromatography screen (Bidet et al., 2014), complex pull-down assay and co-ap assay (Dumrong Mairiang et al., 2013), and colocalization (Balinsky et al., 2013) based PPIs are added in the database to build a high quality PPI repository. All the interactions in the database are annotated with dengue protein, human protein, interaction type, dengue serotype, experimental procedure, article name, author name, and PubMed publication id (PMID) (Supplementary Material S1).

2.1.2. Human protein interaction database

A human protein interaction database is prepared to depict the relationship between the human proteins, present in DenvInt database, with the rest of human proteins in human PPI network. There are many databases, publicly available, to collect and store the human PPI data, such as BioGrid, MINT, HPRD, STRING, and so on. DenvInt database contains 341 distinct human proteins, and all the human proteins may not be equally responsible for pathway analysis. So, to reduce the computational complexity we have considered only those human proteins that are identified by the γ-quasi-biclique algorithm on DenvInt database. Then these human proteins are chosen as input to STRING database (http://string-db.org) to prepare a human PPI database. We have considered four active prediction parameters—Text-mining, Co-Expression, Experiments, and Databases and a high confidence value 0.700 and the minimum threshold for evidence score 0.621 to accumulate the PPI information.

2.1.3. Human protein-disease association database

The human protein-disease association database is downloaded from http://geneticassociationdb.nih.gov. The database is prepared from the published articles based on complex diseases and disorders. The database contains ∼12,400 interactions, including 4260 unique diseases and 3576 unique human proteins/genes. The human-disease association database is given in Supplementary Material S2.

2.2. Definitions

Before discussing the proposed methods and algorithms, some terms are needed to be known.

Bipartite graph: A bipartite graph is a graph whose vertices can be divided into two disjoint sets. Let U and V be the two independent sets of vertices of the graph G. Then every edge (u, v) either connects a vertex from U to V or a vertex from V to U. No edge will connect vertices of the same set.

Biclique: A biclique is a complete bipartite subgraph, where every vertex of one set is connected to every vertex of the second set. Let G = (X ∪ Y, E) be the bipartite graph, having two sets of vertices X and Y and E be the set of edges of the graph. A bipartite subgraph of G, denoted as G′ = (X′ ∪ Y′, E′), X′ ⊆ X, Y′ ⊆ Y, E′ ⊆ E, is called a biclique if every vertex of X′ is connected with every vertex of Y′ and vice-versa, that is, |E′| = |X′| × |Y′|.

Clique: It is a subset of vertices such that its induced subgraph is complete. Let the graph be G = (V, E) where V is the set of vertices and E is the set of edges. A subgraph of G, denoted as G′ = (V′, E′), V′ ⊆ V, and E′ ⊆ E, is called a clique if any two vertices of V′ are connected with each other, that is, $| E' | = (\begin{matrix} | V' | \\ 2 \end{matrix})$ .

Quasi-biclique: A quasi-biclique is an almost complete bipartite subgraph with a fixed density. Let G = (X ∪ Y, E) be the bipartite graph, having two sets of vertices X and Y, and E be the set of edges of the graph. A bipartite subgraph of G, denoted as G′ = (X′ ∪ Y′, E′), X′ ⊆ X, Y′ ⊆ Y, E′ ⊆ E, is called a δ-quasi-biclique, if its density is greater or equal to δ, that is, $\frac{| E' |}{| X' | \times | Y' |} \geq δ$ , where 0 ≤ δ ≤ 1.

Quasi-clique: A quasi-clique is an almost complete subgraph with some fixed density. In some applications, quasi-clique is more suitable, rather than a clique. Let the given graph be G = (V, E) and the fixed density be γ, where 0 ≤ γ ≤ 1. The subgraph G′ = (V′, E′), V′ ⊆ V, E′ ⊆ E is called a γ-quasi clique if its density is greater than or equal to γ, that is, $\frac{| E' |}{(\begin{matrix} | V' | \\ 2 \end{matrix})} \geq γ$ .

3. Algorithms

3.1. Quasi-biclique algorithm

A biclique is a complete bipartite subgraph. It consists of two disjoint vertex subsets where every vertex is adjacent to all vertices in the other subset (Li et al., 2008). In many studies, the quasi-biclique algorithm has been used to predict the PPI, interaction sites, and motifs (Liu et al., 2008, 2010). All PPI related data are not usually fully available. So, the subgraphs created by interacting protein group pairs may not always perfect bicliques (Liu et al., 2008). They are more often near-complete bipartite subgraphs. Therefore, if we try to find bicliques, some important protein interactions may be missed. For this reason, we apply quasi-biclique algorithm rather than bicliques so that the interactions which are absent in a protein interaction subnetwork can be discovered.

In this study, we propose a new technique to generate quasi-bicliques from the PPI network (Algorithm 1).

Algorithm 1: Algorithm for γ Quasi-biclique

Input: Dengue-Human PPI network in the form of Graph G = (V, E)

Output: γ-quasi biclique

1. Biclique generation: First, all the maximal bicliques from the dengue-human PPI database (DenvInt) are generated by edge maximum biclique algorithm (Zhang et al., 2014) where one set represents dengue proteins and other set human proteins. It has been seen that for biological applications, the edge maximum biclique algorithm generates more balanced network compared to the vertex maximum biclique algorithm. To reduce a large number of biclique creations, the minimum number of dengue proteins in the set is kept as 2. Let the bipartite graph be G = (P, E), the vertex set P can be divided into two nonempty sets P1, Q1, and every vertex in P1 (human proteins) directly links to every vertex in Q1 (dengue proteins), where P1 ⊆ P, Q1 ⊆ P, and |E| = |P1| × |Q1|.

2. Quasi-biclique generation: Choose a human protein from the dataset. Add the human protein to the bicliques if it is not present in that biclique and if it interacts with at least one dengue protein in that biclique. For example, let the biclique contain dengue proteins C, E, and NS1 and a set of human proteins (H). The density of the bicliques is 1. Now, a human protein, say, Pr, is randomly chosen from the dengue-human PPI dataset. This Pr will be added to the biclique if the human protein set does not contain Pr and if it interacts with at least one dengue proteins of the corresponding biclique. The algorithm of generating quasi-bicliques iterated for 341 times for 341 distinct human proteins is present in the dengue-human PPI.

3. Density calculation: After all quasi-bicliques are created, density of each of them is calculated considering number of vertices and the number of edges. Let the quasi-biclique be Q = (U1 ∪ V1, E′), where U1 denotes the dengue protein set and V1 denotes the human protein set. The density, γ, of the quasi-bicliques can be calculated as = n (E′)/n (U1) ^* n (V1), where, n (E′) is the number of edges in the graph, n (U1) and n (V1) are the number of nodes in each set. For example, in the Figure 1b, the n (U1) is 2 and n (V1) is 4. So, the number of possible edges is n (U1) ^* n (V1) = 2 ^* 4 = 8. But the quasi-biclique contains five edges. So, the density is (5/8) = 0.625. For γ = 1, the above bipartite graph becomes biclique. In our algorithm, we have used the threshold for the density, γ, of the network 0.5 to generate moderate density quasi-bicliques.

4. Redundant quasi-biclique removal: As some quasi-bicliques contain similar vertex set, the algorithm removes the redundant and spurious ones and generates the maximal quasi-bicliques.

3.2. Quasi-clique algorithm

Cliques are the largest complete subgraphs within a given graph (Brunato et al., 2007). Now, maximum size subgraphs that are almost fully connected are called quasi-cliques. We have developed the quasi-clique algorithm based on the degree of the proteins of the human PPI network. We have applied the heuristic approach to finding the all possible quasi-cliques of the PPI graph.

FIG. 1.

(a) Biclique with γ = 1, (b) γ-quasi biclique with γ = 0.625.

Let G = (V, E) is a graph, where V and E are vertex and edge set of the graph, respectively. Let SU be the subgraph of G. According to the clique relaxation model, SU will be called as quasi-clique if the degree of each vertex is at least γ (|SU| − 1), where γ signifies the density of the network (Pastukhov et al., 2018). The algorithm can be described as follows (Algorithm 2):

Algorithm 2: Algorithm for γ quasi-clique

Input: Human PPI network in form of Graph G = (V, E)

Output: γ-quasi clique

1. Choose a vertex (human protein) randomly from the PPI network.

2. Calculate the degree of the vertex. If it is smaller than the minimum degree (γ (|SU| − 1)), then the vertex is removed. Otherwise, the vertex is added to the graph.

3. Iterate the above two steps (1 to 2) until all vertices (human proteins) of the PPI network are visited.

4. A large number of quasi-cliques are generated by the algorithm. Most of the literature considers the density of the quasi-cliques, γ ≥ 1/2 to form “dense subgraphs” (Pastukhov et al., 2018). We have also set the lower range of γ value 0.5, and upper range less than 1 as γ = 1 signifies cliques. So, the populated quasi-cliques that do not meet the criteria are removed. Having one or two number of vertices in the quasi-cliques do not have any biological significance. So, all the quasi-cliques having less than three vertices are also eliminated.

4. Proposed Methodology

The proposed methodology of finding gateway proteins of DENV infection is described in this section. Most of the biological networks are incomplete due to the lack of experimental data. The incompleteness of these networks is modeled as a quasi-biclique and quasi-clique problem. Two sets of gateway proteins are identified in this study. The steps of generating gateway proteins are described below.

We have prepared dengue-human PPI database from literature and some existing databases. This interaction is modeled as a bipartite network where dengue proteins and human proteins are present in two different sets.

Then the γ-quasi-biclique algorithm, proposed above, is applied to this bipartite graph to gather strong interactions between dengue protein and human protein. We have varied the density of the network γ from 0.5 to 0.9, and the minimum number of dengue proteins considered in a quasi-biclique is 2.

Using this algorithm we got four quasi-bicliques: QB1, QB2, QB3, and QB4. The densities and number of dengue and human proteins of these quasi-bicliques are shown in Table 1. The second quasi-biclique QB2 consists of 4 dengue proteins, C, E, NS3, and NS5, and 54 human proteins. These four dengue proteins are the top four highest degree dengue proteins in the dengue–human PPI network.

Now unique human proteins of these quasi-bicliques are fed to String database to prepare a human PPI. It depicts the interaction between dengue-focused human proteins with the rest of the human proteins.

Then the γ-quasi-clique algorithm is applied to this human PPI to identify the human proteins that overlap with the human proteins generated by γ-quasi-biclique in step 2. We have considered the range of γ value from 0.5 to 0.9. By means of this procedure, we got eight quasi-cliques as shown in Table 2. The corresponding quasi-bicliques and densities are also mentioned in Table 2. From the human proteins present in the QB1 we got QC1, from QB2 we found QC2, from QB3 we found three quasi-cliques-QC3, QC4, and QC5, and from QB4 we got three quasi-bicliques-QC6, QC7, and QC8.

Some of the human proteins of each of the eight quasi-cliques overlap with the four quasi-bicliques. These overlapping human proteins are considered as gateways of infection. The list of overlapping proteins and corresponding quasi-cliques and quasi-bicliques is stated in Table 3. These overlapping human proteins are further analyzed to predict its functionality.

The human gene-disease association database is downloaded from http://geneticassociationdb.nih.gov A bipartite graph is prepared where human proteins are on one side and different diseases are on another side of the graph. The edges of the graph signify the association of human proteins with diseases. Then, the above described γ quasi-biclique algorithm is applied to this network. Instead of dividing the bipartite graph into dengue protein set and human protein set, here it is divided into human proteins in one set and all the diseases in the other set. Three Quasi-bicliques are generated, QBD1, QBD2, and QDB3, considering density γ threshold 0.5 (Table 4). The overlapping human proteins between these quasi-bicliques and quasi-cliques are reported in Table 4. These proteins are very important for any type of disease spreading as they are gateway proteins.

Table 2.

Quasi-Cliques Found from Human Protein Interaction Database That Overlap with the Human Proteins Present in the Quasi-Bicliques

Quasi-cliques	Corresponding quasi-bicliques	Human proteins	Density of the quasi-cliques
QC1	QB1	Count: 68 FAU PSMA1 PSMB4 PSMC1 PSMC2 PSMC3 PSMC4 PSMC5 PSMC6 PSMD1 PSMD11 PSMD12 PSMD3 RPL11 RPL12 RPL13 RPL14 RPL15 RPL18 RPL18A RPL19 RPL21 RPL23 RPL23A RPL24 RPL27 RPL27A RPL29 RPL30 RPL31 RPL32 RPL35 RPL36 RPL37A RPL38 RPL4 RPL5 RPL6 RPL7 RPL8 RPL9 RPS10 RPS11 RPS12 RPS13 RPS14 RPS15 RPS15A RPS16 RPS19 RPS2 RPS20 RPS21 RPS23 RPS24 RPS26 RPS27 RPS27A RPS28 RPS29 RPS3 RPS3A RPS4X RPS5 RPS6 RPS7 RPS8 RPSA	0.71
QC2	QB2	Count: 80 FAU GTPBP4 HSP90AB1 NFKBIA PSMA1 PSMA6 PSMB2 PSMB4 PSMC1 PSMC2 PSMC3 PSMC4 PSMC5 PSMC6 PSMD1 PSMD11 PSMD12 PSMD13 PSMD14 PSMD2 PSMD3 PSMD4 PSMD6 PSMD7 PSMD8 RPL11 RPL12 RPL13 RPL14 RPL15 RPL18 RPL18A RPL19 RPL21 RPL23 RPL23A RPL24 RPL27 RPL27A RPL30 RPL31 RPL32 RPL35 RPL37A RPL38 RPL4 RPL5 RPL6 RPL7 RPL8 RPL9 RPS10 RPS11 RPS12 RPS13 RPS14 RPS15 RPS15A RPS16 RPS17 RPS19 RPS2 RPS20 RPS21 RPS23 RPS24 RPS26 RPS27 RPS27A RPS28 RPS29 RPS3 RPS3A RPS4X RPS5 RPS6 RPS7 RPS8 RPSA WWP1	0.544
QC3	QB3	Count: 18 BOD1L1 CLU CTNNB1 DIABLO EEF2 EGFR EIF4A1 EIF4G2 FAU GNB2L1 HNRNPC HSP90B1 MAP3K7 NAP1L1 PAIP1 PTBP1 RPL13 RPL27A	0.562
QC4	QB3	Count: 10 BIRC2 DIABLO MAP3K7 RPS27A TAB1 TAB2 TRAF1 TRAF2 TRAF6 UBC	0.644
QC5	QB3	Count: 5 CTNNB1 EGFR RPS27A STAT3 UBC	0.7
QC6	QB4	Count:8 EGFR HSP90B1 IL10RA JAK1 JAK2 SRC STAT3 UBC	0.75
QC7	QB4	Count:8 CALR EGFR HSP90B1 HSPA5 PDIA3 PRKCSH STAT3 UBC	0.75
QC8	QB4	Count: 6 IL10 IL10RA JAK1 JAK2 STAT3 UBC	0.86

Table 3.

Overlapping Human Proteins of Quasi-Cliques and Quasi-Bicliques

Quasi-cliques	Overlapping proteins with the quasi-bicliques (gateway proteins)	Corresponding quasi-bicliques
QC1	PSMC1	QB1, QB2
QC1	RPL12	QB1, QB2
QC1	RPL24	QB2
QC1	RPL6	QB1, QB2
QC1	RPL5	QB1, QB2
QC1	RPS27	QB1, QB2, QB3
QC1	RPS7	QB1, QB2
QC2	GTPBP4	QB1, QB2
QC2	HSP90AB1	QB2
QC2	NFKBIA	QB2
QC2	PSMC1	QB1, QB2
QC2	RPL12	QB1, QB2
QC2	RPL24	QB1, QB2
QC2	RPL5	QB1, QB2
QC2	RPL6	QB1, QB2
QC2	RPS27	QB1, QB2, QB3
QC2	RPS7	QB1, QB2
QC2	WWP1	QB1, QB2
QC3	BIRC2	QB1, QB2, QB3
QC4	BIRC2	QB1, QB2, QB3
QC4	CLU	QB3
QC4	HNRNPC	QB3
QC4	NAP1L1	QB1, QB2, QB3
QC4	PAIP1	QB3
QC4	PTBP1	QB3
QC4	RPS27	QB1, QB2, QB3
QC4	RPS7	QB1, QB2
QC4	RRP12	QB1, QB2, QB3
QC4	STAT3	QB3, QB4
QC5	STAT3	QB3, QB4
QC6	STAT3	QB3, QB4
QC7	STAT3	QB3, QB4
QC7	CALR	QB1, QB3, QB4
QC7	HSPA5	QB1, QB4
QC8	STAT3	QB3, QB4
QC8	IL10	QB4

Table 4.

Quasi-Bicliques Generated from Human Protein-Disease Association Network Corresponding to Quasi-Cliques

Quasi-biclique	Corresponding QC	Overlapping human proteins (gateway proteins)	Diseases	Density
QBD1	QC3, QC5	Count: 2 STAT3, EGFR	Asthma, kidney cancer, colorectal cancer, multiple sclerosis	0.625
QBD2	QC4	Count: 4 STAT3, IL10, IL10RA, JAK2	Asthma, Crohn's disease, HIV, infertility, tubal factor, inflammation—premature birth, schizophrenia, ulcerative colitis	0.867
QBD3	QC6	Count: 4 EGFR, STAT3, JAK2	Asthma, Crohn's disease	0.666

So, we have explored three networks in this article. They are dengue–human interaction network, human protein interaction network, and human protein-disease association network to find the potential pathways of infection by the DENVs that lead to various diseases in the human body. We have generated four quasi-bicliques from dengue–human interaction network, eight quasi-cliques from human protein interaction network, and three quasi-bicliques from human protein-disease association network. The overlapping relation among them is demonstrated in the following Figure 2.

FIG. 2.

The diagrammatic representation of the overlapping relation between quasi-bicliques and quasi-cliques. The overlapping proteins are treated as gateway proteins.

5. Results and Discussion

5.1. KEGG pathway analysis of quasi-cliques

We have analyzed the KEGG pathways and GO of the quasi-cliques using the DAVID tool. Few significant GO terms along with KEGG pathways of the four quasi-cliques are stated in Table 5. QC1 mainly consists of the proteins that participate in translational elongation biological process. This quasi-clique QC1 overlaps with QB1, QB2, and QB3. The list is given in Table 5. Now human proteins PSMC1, RPL12 of QC1 and QC2 overlap with QB1 and QB2 and they also interact with dengue proteins NS3, NS5, C, and E. In the same manner, RPL24 of QC1 overlaps with QB2 and interacts with dengue protein NS3 and E. We have examined that QC1, QC2, and QC3 the three quasi-cliques have same KEGG pathway ribosome. Now, the ribosome is one of the important cellular units responsible for making proteins. The same ribosomal proteins (RPS7, RPS27, and RPL12) are present in all these quasi-cliques. Besides this, QC1 and QC2 contain three other ribosomal proteins—RPL5, RPL6, and RPL24.

Table 5.

The Significant Important Gene Ontology Terms and KEGG Pathways Found in the Quasi-Cliques

Quasi-cliques	Biological process	Significant GO term molecular function	Cellular component	KEGG pathways
QC1	Translational elongation (p = 6.7E-113)	Structural constituent of ribosome (p = 2.6E-97)	Cytosolic ribosome (p = 1.3E-100)	Ribosome (p = 1.1E-92)
QC2	Translational elongation (p = 7.6E-103)	Structural constituent of ribosome(54) (p = 1.6E-89)	Cytosolic ribosome (p = 4.5E-92)	Ribosome (p = 1.6E-83)
QC3	Translational elongation (p = 2.7E-65)	Structural constituent of ribosome (p = 1.2E-47)	Cytosolic small ribosomal subunit (p = 7.6E-68)	Ribosome (p = 4.3E-47)
QC4	Regulation of apoptosis (p = 8.3E-8)	Zinc ion binding (p = 6.4E-4)	Cytosol (p = 1.3E-8)	NOD-like receptor signaling pathway (p = 1.4E-6)
QC4	Regulation of programmed cell death (p = 8.9E-8)	Metal ion binding (p = 1.9E-3)	Cytosol (p = 1.3E-8)	Small cell lung cancer (p = 2.3E-4)
QC5	Positive regulation of nitrogen compound metabolic process (p = 5.1E-6)	Transcription regulator activity (p = 5.8E-3)	Cytosolic part (p = 8.3E-4)	Pathways in cancer (p = 2.3E-2)
QC6	Enzyme linked receptor protein signaling pathway (bp-FAT) (p = 5.8E-6)	Protein tyrosine kinase activity (p = 6.9E-5)	Caveola (p = 2.8E-2)	Jak-STAT signaling pathway (p = 8.9E-4)
QC7	Regulation of apoptosis (p = 1.4E-5)	Calcium ion binding (p = 7.3E-4)	Endoplasmic reticulum lumen (p = 8.1E-6)	Antigen processing and presentation (p = 3.8E-3)
QC8	Protein kinase cascade (p = 8.0E-5)	Growth hormone receptor binding (p = 1.9E-3)	Nonmembrane-bounded organelle (p = 6.0E-2)	Jak-STAT signaling pathway (p = 4.1E-6)

GO, gene ontology.

Several studies showed that ribosomal biogenesis is evidently related to cancer especially colorectal cancer (Lempiäinen and Shore, 2009; Lascorz et al., 2011). It also has direct influence in tumor genesis through some ribosomal functions (Mao-De and Jing, 2007). Thus, DENV can lead to or cause several tumors and cancer through these ribosomal gateway proteins—RPS7, RPS27, RPL12, RPL5, RPL6, and RPL24. The 10 proteins of quasi-clique QC4 regulate the apoptosis, and the KEGG pathway is NOD-like receptor signaling pathway and small cell lung cancer. The GO molecular function is zinc ion binding. There is only one human protein BIRC2 of QC4 that overlaps with all the three quasi-bicliques QB1, QB2, and QB3. BIRC2 interacts with dengue proteins NS5, NS1, and C. So, these dengue proteins can damage the Nod-like receptor signaling pathway and cause lung cancer by the gateway protein BIRC2.

QC5 mainly consists of the proteins that participate in the positive regulation of nitrogen compound metabolic process and transcription regulator activity. So, malfunctioning of these proteins may hamper the normal regulatory roles of these transcription factors. These proteins are also involved in the KEGG pathways in cancer. Human protein STAT3 of QC5 overlaps with quasi-biclique QB3 and also interacts with dengue proteins E and NS1. So, this study suggests that dengue proteins E and NS1 may show the way to cancer in the human body, and the main gateway protein responsible for that is STAT3. Some interesting evidence shows the relationship between mosquito-borne diseases and cancer in Benelli et al. (2016). The cancer pathways are activated by mosquito biting. For example, Aedes aegypti which carries DENV transfers reticulum sarcoma with the help of tumor cells.

The host proteins of QC6 and QC8 are involved in a JAK-STAT signaling pathway, which signifies that the dengue proteins in QB4 interact with QC6 and QC8 through the common human protein STAT3. Now, this JAK-STAT signaling pathway transmits cellular signals to the nucleus of the cells, which handles DNA transcription of the cell. So, disruption of JAK-STAT functionality may cause different immune deficiency syndromes and cancers (Ivashkiv and Hu, 2004). Different studies (Lin et al., 2004; Souza-Neto et al., 2009) have identified that when dengue proteins interact with human proteins, they suppress the JAK-STAT signaling pathway. The quasi-clique QC7 consists of eight human proteins, which involve in calcium ion binding and regulate the apoptosis. The KEGG pathway associated with these proteins is antigen processing and presentation. By this technique, the cells assimilate proteins from inside or outside the cell. The gateway proteins that are responsible for this activity are STAT3, HSPA5, and CALR. The association of dengue proteins and antigen processing is reported in Hershkovitz et al. (2008).

So, GO and KEGG pathway analysis of the eight quasi-cliques helps to identify the gateway proteins, which are captured by the dengue proteins at the time of disease spreading in the human body. Dengue proteins may damage the regulatory system and immune system and may lead to different types of cancers, including lung cancer, colorectal cancer, and so on.

5.2. Literature support of the diseases predicted by quasi-cliques and human-disease association network

To analyze the diseases associated with the human proteins in the quasi-cliques for getting the possible pathway of pathogenesis leading to various diseases, we have applied the quasi-biclique algorithm on the human gene-disease association network corresponding to eight quasi-cliques. We got three quasi-bicliques—QBD1, QBD2, and QBD3 as shown in Table 4. QBD1 overlaps with two quasi-cliques—QC3 and QC5 with two human proteins, namely, STAT3 and EGFR. These two proteins are associated with four different diseases. QBD2 overlaps with QC4 with four proteins, which are connected with seven diseases. QBD3 overlaps with QC6 with three host proteins, which are associated with two diseases.

We have searched existing literature in support of the diseases that are reported in Table 4. Many of them are already established in recent articles. Asthma is a chronic long-term lung disease that narrows the airways of the lung. In Pang et al. (2017), the authors identified diabetes, cardiac disorders, and asthma, as independent risk factors of severe organ involvement during dengue infection. DENV causes colitis and other liver and kidney diseases (Park et al., 2008; Lizarraga and Nayer, 2014; Samanta and Sharma, 2015). DENV infection during pregnancy causes multiple problems like miscarriage, stillbirth, infertility, premature birth, and so on. (Ribeiro et al., 2017). It may also cause bleeding tendencies if she goes in for cesarean section (Chitra and Panicker, 2011). Multiple sclerosis is a central nervous system disease that disrupts the flow of information within the brain. Recently published literature has proved that some dengue fever causes multiple sclerosis as blood cells are affected by dengue fever (Fragoso and Brooks, 2016).

Schizophrenia is one type of mental disorder defined by abnormal social behavior. Research shows that as DENV affects the neurons, it increases the risk of schizophrenia in the patients. An 18-year-old male student from a rural family was diagnosed with schizophrenia-like disorder due to dengue viral fever in Fragoso and Brooks (2016)). Dengue infection increases the level of TNF-α, which may stimulate inflammation and endothelial dysfunction (Kurane, 2007; Cheng et al., 2015). There are lots of studies that are reported to identify the dengue-HIV coinfection issue. Some research explores that dengue protein NS5 downregulates HIV coreceptor protein CXCR4 (López-Lemus et al., 2012; Pang et al., 2015). Six clinical reports of dengue serotype-1 infection in HIV infected patients are identified in Delgado-Enciso et al. (2017). Case studies showed that DENV-HIV coinfected patients have higher eosinophil proportion and pulse rate but lower serum hematocrit level (Pang et al., 2015). According to the WHO 2009 criteria, this coinfection causes a major problem due to anemia and severe plasma leakage. So, these types of patients need high monitoring from the first stage.

The above analysis indicates that most of the diseases found in Table 5 have evidence in published articles for their connection with DENV transmission. The DENV affects the gateway proteins, namely, EGFR, STAT3, IL10, CALR, JAK2, and IL10RA to cause these diseases in the human body. So, the algorithms we have used in this study may help to get possible pathways of DENV infection leading to various diseases.

5.3. Gateway protein analysis

In this study, we have identified two sets of gateway proteins. Using the KEGG pathway and GO analysis process we got 22 gateway proteins for dengue infection. They are PSMC1, RPS7, WWP1, RPS27, RPL12, RPL5, RPL6, RPL24, RRP12, CLU, HNRNPC, NAP1L1, PAIP1, PTBP1, STAT3, HSPA5, BIRC2, GTPBP4, HSP90AB1, NFKBIA, IL10, and CALR. The second set contains six gateway proteins EGFR, STAT3, IL10, CALR, JAK2, and IL10RA using the quasi-biclique algorithm on human protein-disease association network. These proteins play a very significant role in dengue infection in the human body. The average degree of the gateway proteins is significantly high (26.56) compared to nongateway proteins (7.94). The degrees (calculated from HPRD database release 9) to these gateway proteins are given in Supplementary Material S3. This degree difference may reflect the fact that viral proteins tend to attack the proteins having higher connectivity so that the maximum number of neighbor proteins are affected and, hence, causes multiple other diseases in the human body. It is evident that most of the gateway proteins of the first set are involved in JAK-STAT signaling pathways and pathways in cancer. Gateway protein BIRC2 is directly related to esophageal cancer (Brown et al., 2011) and cervical cancer (Choschzick et al., 2012). We have predicted that a list of ribosomal proteins, such as RPS27, RPL12, RPL5, RPL6, RPL24, RRP12, and RPS7, is treated as infection gateway by DENV.

In Wang et al. (2015), the authors established the fact that ribosomal proteins are linked to the development of different hematological, metabolic, and cardiovascular diseases and cancer. So, dengue infection can lead to several ribosome related diseases. HSPA5 is an important protein, which regulates ER stress response in the human cell. All the viral proteins specially capsid proteins of dengue, Ebola, influenza, etc., synthesize a huge amount of proteins and induce an ER stress response in the infected cells by stimulating HSPA5 expression (Booth et al., 2015).

The second set is concerned with diseases such as asthma, ulcerative colitis, multiple sclerosis, premature birth, kidney, and colorectal cancer. The degree of EGFR protein in the HPRD database (version 9) is 163. So, a large number of proteins can be affected by the human body through the EGFR protein network. In Komposch and Sibilia (2016), the authors illustrated that a change in EGFR protein's signaling can cause acute and chronic liver damage. In addition, the deregulation of EGFR signaling pathway causes chronic lung disease and also lung cancer (Vallath et al., 2014). Rest of the diseases associated with these gateway proteins are reported in Supplementary Material S4. According to our analysis, dengue proteins attack the gateway proteins to enter into the human PPI network. So, the probability of the occurring diseases caused by these gateway proteins increases during or after the dengue infection in the human body.

5.4. Direct PPI with diseases

The dengue–human PPI database contains 545 interactions between 10 dengue proteins and 341 unique human proteins. We have prepared the database from published literature. The human protein-disease association database contains ∼12,400 associations, including 4260 unique diseases and 3576 unique human proteins. Now among the 341 unique human proteins of dengue-human PPI, 88 human proteins overlap with the human-disease database. Some of these proteins are associated with more than one disease. The detailed list is given in Supplementary Material S5. The degree of each human protein is depicted in Figure 3. Human protein APOE (apolipoprotein E) has maximum degree 88, that is, APOE is associated with 88 different types of diseases. There are 45 human proteins which have degree 1 as they are responsible for one type of disease.

FIG. 3.

Frequency of dengue viral interacting human proteins with different human diseases.

6. Conclusion

In this study, we have identified the possible infection pathway of DENV in the human body. For this reason, first, we have calculated the quasi-bicliques of dengue-human PPI. Then the proteins of these quasi-bicliques are mapped onto the quasi-cliques of the human PPI. The quasi-cliques are also used to find the overlapping proteins with human protein-disease association networks. So, these two sets of overlapping proteins are described as gateway proteins for DENV infection. They are very important in terms of disease spreading in the human body as dengue proteins attack these gateway proteins to enter into the human PPI network. We have analyzed the KEGG pathway and GO of these gateway proteins. This analysis reveals that these proteins are related to different diseases such as lung cancer, kidney cancer, asthma, ulcerative colitis, multiple sclerosis, premature birth, infertility, and so on. Most of the predicted diseases have evidence in the published literature.

Footnotes

Acknowledgments

The authors thank their entire group members for helping them.

Author Disclosure Statement

The authors declare they have no competing financial interests.

Funding Information

The authors received no financial support for the research of this article.

Supplementary Material

References

Balinsky

C.A.

, Schmeisser

, Ganesan

, et al. 2013. Nucleolin interacts with the dengue virus capsid protein and plays a role in formation of infectious virus particles. J. Virol. 87, 13094–13106.

Benelli

, Iacono

A.L.

, Canale

, et al. 2016. Mosquito vectors and the spread of cancer: An overlooked connection?. Parasitol. Res. 115, 2131–2137.

Bidet

, Dadlani

, and Garcia-Blanco

M.A.

2014. G3BP1, G3BP2 and CAPRIN1 are required for translation of interferon stimulated mRNAs and are targeted by a dengue virus non-coding RNA. PLoS Pathog. 10, e1004242.

Booth

, Roberts

J.L.

, and Dent

2015. HSPA5/Dna K may be a useful target for human disease therapies. DNA Cell Biol. 34, 153–158.

Brown

, Bothma

, Veale

, et al. 2011. Genomic imbalances in esophageal carcinoma cell lines involve Wnt pathway genes. World J. Gastroenterol. 17, 2909–2923.

Brunato

, Hoos

H.H.

, and Battiti

2007. On effectively finding maximal quasi-cliques in graphs, 41–55. In International Conference on Learning and Intelligent Optimization. Springer, Trento, Italy.

Cheng

Y.-L.

, Lin

Y.-S.

, Chen

C.-L.

, et al. 2015. Dengue virus infection causes the activation of distinct NF-κB pathways for inducible nitric oxide synthase and TNF-α expression in RAW264.7 cells. Mediators Inflamm. 2015, 274025.

Chitra

T.V.

, and Panicker

2011. Maternal and fetal outcome of dengue fever in pregnancy. J. Vector Borne Dis. 48, 210.

Choschzick

, Tabibzada

A.M.

, Gieseking

, et al. 2012. Birc2 amplification in squamous cell carcinomas of the uterine cervix. Virchows Arch. 461, 123–128.

10.

Chowdhary

, Zhang

, and Liu

J.S.

2009. Bayesian inference of protein–protein interactions from biological literature. Bioinformatics, 25, 1536–1542.

11.

Delgado-Enciso

, Espinoza-Gómez

, Ochoa-Jiménez

, et al. 2017. Dengue infection in a human immunodeficiency virus-1 positive patient chronically infected with hepatitis B virus in Western Mexico. Am. J. Trop. Med. Hyg. 96, 122–125.

12.

Dey

, and Mukhopadhyay

2017. Denvint: A database of protein–protein interactions between dengue virus and its hosts. PLoS Negl. Trop. Dis. 11, e0005879.

13.

Doolittle

J.M.

, and Gomez

S.M.

2011. Mapping protein interactions between dengue virus and its human and insect hosts. PLoS Negl. Trop. Dis. 5, e954.

14.

Folly

B.B.

, Weffort-Santos

A.M.

, Fathman

C.G.

, et al. 2011. Dengue-2 structural proteins associate with human proteins to produce a coagulation and innate immune response biased interactome. BMC Infect. Dis. 11, 34.

15.

Fragoso

Y.D.

, and Brooks

J.B.B.

2016. Encephalomyelitis associated with dengue fever. JAMA Neurol. 73, 1368–1368.

16.

Hershkovitz

, Zilka

, Bar-Ilan

, et al. 2008. Dengue virus replicon expressing the nonstructural proteins suffices to enhance membrane expression of HLA class I and inhibit lysis by human NK cells. J. Virol. 82, 7666–7676.

17.

Ivashkiv

L.B.

, and Hu

2004. Signaling by stats. Arthritis Res. Ther. 6, 159.

18.

Khadka

, Vangeloff

A.D.

, Zhang

, et al. 2011. A physical interaction network of dengue virus and human proteins. Mol. Cell. Proteomics 10, M111.012187.

19.

Komposch

, and Sibilia

2016. EGFR signaling in liver diseases. Int. J. Mol. Sci. 17, 30.

20.

Kurane

2007. Dengue hemorrhagic fever with special emphasis on immunopathogenesis. Comp. Immunol. Microbiol. Infect. Dis. 30, 329–340.

21.

Lascorz

, Hemminki

, and Försti

2011. Systematic enrichment analysis of gene expression profiling studies identifies consensus pathways implicated in colorectal cancer development. J. Carcinog. 10, 7.

22.

Le Breton

, Meyniel-Schicklin

, Deloire

, et al. 2011. Flavivirus NS3 and NS5 proteins interaction network: A high-throughput yeast two-hybrid screen. BMC Microbiol. 11, 234.

23.

Lempiäinen

, and Shore

2009. Growth control and ribosome biogenesis. Curr. Opin. Cell Biol. 21, 855–863.

24.

, Sim

, Liu

, et al. 2008. Maximal quasi-bicliques with balanced noise tolerance: Concepts and co-clustering applications, 72–83. In Proceedings of the 2008 SIAM International Conference on Data Mining. SIAM, Atlanta, GA, USA.

25.

Lin

R.-J.

, Liao

C.-L.

, Lin

, et al. 2004. Blocking of the alpha interferon-induced JAK-STAT signaling pathway by Japanese encephalitis virus infection. J. Virol. 78, 9285–9294.

26.

Liu

H.-B.

, Liu

, and Wang

2008. Searching maximum quasi-bicliques from protein-protein interaction network. J. Biomed. Sci. Eng. 1, 200.

27.

Liu

, Li

, and Wang

2010. Modeling protein interacting groups by quasi-bicliques: Complexity, algorithm, and application. IEEE/ACM Trans. Comput. Biol. Bioinform. 7, 354–364.

28.

Lizarraga

K.J.

, and Nayer

2014. Dengue-associated kidney disease. J. Nephropathol. 3, 57.

29.

López-Lemus

U.A.

, Valle-Reyes

, Ochoa-Jiménez

, et al. 2012. HIV-related problems: 460 detection of dengue virus in HIV-1-infected patients during the epidemic in 2009 in Colima, Mexico. World Allergy Organ. J. 5(Suppl 2), S163.

30.

Mairiang

2012. Characterization of Intracellular Interactions Between Dengue Virus and Host Proteins. Wayne State University, Detroit, MI, USA.

31.

Mairiang

, Zhang

, Sodja

, et al. 2013. Identification of new protein interactions between dengue fever virus and its hosts, human and mosquito. PLoS One, 8, e53535.

32.

Mao-De

, and Jing

2007. Ribosomal proteins and colorectal cancer. Curr. Genomics, 8, 43–49.

33.

Pang

, Hsu

J.P.

, Yeo

T.W.

, et al. 2017. Diabetes, cardiac disorders and asthma as risk factors for severe organ involvement among adult dengue patients: A matched case-control study. Sci. Rep. 7, 39872.

34.

Pang

, Thein

T.-L.

, Lye

D.C.

, et al. 2015. Differential clinical outcome of dengue infection among patients with and without HIV infection: A matched case–control study. Am. J. Trop. Med. Hyg. 92, 1156–1162.

35.

Park

S.B.

, Ryu

S.Y.

, Jin

K.B.

, et al. 2008. Acute colitis associated with dengue fever in a renal transplant recipient, 2431–2432. In Transplantation Proceedings, Volume 40. Elsevier, USA.

36.

Pastukhov

, Veremyev

, Boginski

, et al. 2018. On maximum degree-based-quasi-clique problem: Complexity and exact approaches. Networks, 71, 136–152.

37.

Ribeiro

C.F.

, Lopes

V.G.S.

, Brasil

, et al. 2017. Dengue infection in pregnancy and its impact on the placenta. Int. J. Infect. Dis. 55, 109–112.

38.

Samanta

, and Sharma

2015. Dengue and its effects on liver. World J. Clin. Cases, 3, 125.

39.

Souza-Neto

J.A.

, Sim

, and Dimopoulos

2009. An evolutionary conserved function of the jak-stat pathway in anti-dengue defense. Proc. Natl. Acad. Sci. 106, 17841–17846.

40.

Vallath

, Hynds

R.E.

, Succony

, et al. 2014. Targeting EGFR signalling in chronic lung disease: Therapeutic challenges and opportunities. Eur. Respir. J. 44, 513–522.

41.

Wang

, Nag

, Zhang

, et al. 2015. Ribosomal proteins and human diseases: Pathogenesis, molecular mechanisms, and therapeutic implications. Med. Res. Rev. 35, 225–285.

42.

Yamanishi

, Vert

J.-P.

, and Kanehisa

2004. Protein network inference from multiple genomic data: A supervised approach. Bioinformatics, 20(Suppl_1), i363–i370.

43.

Zhang

, Phillips

C.A.

, Rogers

G.L.

, et al. 2014. On finding bicliques in bipartite graphs: A novel algorithm and its application to the integration of diverse biological data types. BMC Bioinformatics, 15, 110.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.06 MB

0.25 MB

0.01 MB

0.02 MB

A Graph-Based Approach for Finding the Dengue Infection Pathways in Humans Using Protein–Protein Interactions

Abstract

1. Introduction

2. Materials and Methods

2.1. Databases and preprocessing

2.1.1. Dengue–human protein interaction database: DenvInt

2.1.2. Human protein interaction database

2.1.3. Human protein-disease association database

2.2. Definitions

3. Algorithms

3.1. Quasi-biclique algorithm

3.2. Quasi-clique algorithm

4. Proposed Methodology

5. Results and Discussion

5.1. KEGG pathway analysis of quasi-cliques

5.2. Literature support of the diseases predicted by quasi-cliques and human-disease association network

5.3. Gateway protein analysis

5.4. Direct PPI with diseases

6. Conclusion

Footnotes

Acknowledgments

Author Disclosure Statement

Funding Information

Supplementary Material

References

Supplementary Material