For the development of a robust and dynamic database, we employed state-of-the-art MERN stack, which comprises of four essential components: (i) MongoDB, (ii) Express.js, (iii) React, and (iv) Node.js. This combination of technologies is flexible, scalable, and highly suitable for web applications. MongoDB serves as a database management system for storing the large amount of biological data. Express.js (backend framework) was utilized to create RESTful APIs and handle hypertext transfer protocol (HTTP) requests. React, a JavaScript library, was implemented to create a user-friendly interface that facilitates an interactive and responsive platform to access the available data and for protein interaction network visualization. Node.js forms the runtime environment for the server, thus ensuring efficient communication between frontend and backend for high-speed data retrieval and processing. The ‘Homepage’ of the database is presented in Figure 1.

The database consists of 22 MPXV strains interacting with human proteins. Publicly available databases and specialized resources were leveraged for protein datasets collection. The proteome (20,299 proteins) of human was obtained from UniProt Swiss-Prot. ² On the other hand, the proteins of MPXV strains were obtained from NCBI. ³ To reduce the sequence redundancy, CD-HIT (Fu et al., 2012) at 100% similarity was implemented on all the downloaded proteomes. The users can access these protein datasets through the “Datasets” page on the database, which will direct them to the actual data source.

Being comprehensive in nature, HuPoxNET is a resource for various functional annotations of the host and pathogen proteins. These include Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, gene ontology (GO), functional domains, subcellular localization, and potential drug targets. GO terms and KEGG pathways for human proteins were obtained from Ensembl. ⁴ For functional domains, we analyzed the proteins with InterProScan (Jones et al., 2014). The subcellular localization information of human proteins was obtained from the Human Protein Atlas ⁵ (Thul and Lindskog, 2018). The database also contains the information about the potential drugs against MPXV infection obtained from DrugBank. ⁶

To implement the sequence-based computational models (interolog and domain) for the prediction of human-MPXV PPIs, we utilized the available gold standard protein–protein interaction databases (BioGRID, DIP, HPIDB, IntAct, MINT, and VirHostNet) and domain-domain interaction (DDI) databases (3did, DOMINE, and IDDI). These databases were extensively filtered to extract the experimentally validated and species-specific (here, human-virus) interactions, while the non-validated interactions were simultaneously discarded. This step was performed to reduce the false positive predictions. In homology-based interolog model, the protein datasets were subjected to BLAST search against the above-mentioned standard databases to predict the interactions. On the other hand, in domain-based model, we employed HMMER (Eddy, 2011) against Pfam as reference database to identify significant domains of the host and pathogen proteins to predict the potential PPIs. Furthermore, Cytoscape.js (Shannon et al., 1971) plugin was employed for enhanced network visualization of the predicted interactome.

Monkeypox disease is caused in humans by different MPXV strains that differ in their evolutionary patterns (Kataria et al., 2023). In line with this, we implemented 22 MPXV strains causing infection in humans. HuPoxNET supports two sequence-based computational models (interolog and domain) for the prediction of PPIs between human and MPXV proteins. These models are integrated in the “Interactome” module of the database (Figure 2), which filters the precalculated interactions based on various parameters selected by the users. In this module, the users can predict the PPIs using the desired model (interolog, domain, or consensus). The users can select the interaction model of choice, followed by the selection of the available standard PPI and/or DDI databases, and subsequently change the BLAST alignment parameters (identity, coverage, and e-value) for both host and pathogen. The “consensus” model allows the user to predict the interactions based on both the interolog and domain model. This model outputs the common interactions predicted by both computational models, thus increasing the reliability of the predicted interactome.

Alternatively, instead of using the whole proteome, the users can also enter the protein IDs or provide a list of protein IDs of either human or MPXV proteins. This will filter the interactions based on the given protein IDs, thus making the prediction specific and time-efficient. Based on the selected MPXV strain, the resulting table consists of the potential interactions between human and MPXV proteins, corresponding interaction pair from the selected standard database(s), experimental method, domain name, InterProScan ID, source database name, PubMed ID, and confidence score (Figures 3A,B). Additionally, the human proteins in the predicted interactions have been linked to PubMed, Human Protein Atlas, and GTEx portal to access gene expression and related data to monkeypox disease, thus enriching the database with various functional annotations. To enhance user experience and provide more protein information, the resulting proteins in the interactions are also externally linked to the respective publicly available biological databases such as UniProt, NCBI, etc. The predicted interactions can also be downloaded in a comma-separated values (csv) file format. To make the search more user-friendly, each human protein in the predicted interactions is linked to its protein annotations from the results page of interactome module. This comprehensive annotation page consists of various functional annotations of proteins such as gene ontology (GO) terms, KEGG pathways, drug targets, etc.

Following the predictions, the user can visualize the predicted interactome, which is implemented using Cytoscape. The network consists of the predicted interactions, whereby the nodes are sized based on the degree of the protein. Higher the protein degree, larger is the node size, and vice versa. The colored edges in the network corresponds to the different gold standard databases selected by the user to predict the interactions (Figure 4). Selecting a node in the network provides the user with more information such as degree of the protein, which is further linked to external resources. The users can also modify the network according to their research question, and download it in PNG format. Additionally, the database allows the user to download the network in JSON format, thus enabling the users to import the network in advanced visualization tools and perform various analysis to extract more useful information.

The understanding of the molecular mechanisms underlying disease infection and subsequent host defense responses rely on the functional annotations of the proteins involved in the interactions (Hong et al., 2019). HuPoxNET serves as the hub of functional annotations of the human and MPXV proteins involved in the interactions. Various computational tools have been developed, which were implemented to predict the different annotations of the interacting proteins in monkeypox disease. These annotations include functional domains of the proteins, GO terms, KEGG pathways, protein subcellular localization, and drug targets, which can be accessed from the ‘Homepage’ of the database.

Gene ontology provide meaningful annotations of the proteins, and are categorized into 3 domains—cellular component, biological process, and molecular function (Harris et al., 2006). The database contains the GO annotations of the human and viral proteins, along with the description and evidence code of the GO term. KEGG pathways and their respective description have also been implemented in the database. Subcellular localization is essential to understand the function of a protein in the cell. In context of PPIs, the identification of protein localization will provide insights into the location where the host and pathogen proteins interact during viral infection and human defense responses, thus aiding in the identification of potential targets for drug development (Pan et al., 2021). Furthermore, the functional domains of the proteins and related information such as protein length, InterPro ID, domain ID, and description of the domain has been made available for human and MPXV proteins. The potential drug targets for human proteins can also be accessed on the database, which direct to DrugBank for additional information on the drugs. For robust analysis of potential human proteins, these drug targets have been further linked to ChEMBL database ⁷ (Figure 5), which is a comprehensive repository of bioactive drug-like molecules and contains information regarding their biological activities, mechanisms of action, and pharmacological properties (Gaulton et al., 2012).

The information from the above-mentioned protein functional annotations, wherever possible, has been linked to different biological databases. Furthermore, a search bar has been implemented on all the pages that allows the user to query a specific keyword such as kinase, ATP, etc. This makes the database highly efficient and easy to search information.

To increase reliability of the predicted interactions, we identified the validated drugs against monkeypox disease in the literature and searched those in the predicted interactome. Several studies have reported different drugs that were administered (orally, intravenous, or cutaneous) to the patients infected with MPXV, resulting in complete recovery from the disease. These Food and Drug Administration (FDA)-approved antiviral drugs include brincidofovir (Adler et al., 2022), tecovirimat (Almehmadi et al., 2022; Frenois-Veyrat et al., 2022; Matias et al., 2022), cidofovir (Tarín-Vicente et al., 2022; Srivastava et al., 2023), ribavirin (Baker et al., 2003) etc. that are used to treat poxviruses (Ortiz-Saavedra et al., 2022). Another study reported tigecycline as a therapeutic drug against MPXV infection by implementing structure-based virtual screening and molecular simulation (Alandijany et al., 2023). We also identified potential human targets for the above-mentioned drugs (Table 1).

Furthermore, we randomly selected a protein interaction pair (“Q92934-AAY97628.1”) from the predicted interactions in HuPoxNET and searched for the respective human and viral protein annotations in the database. The functional annotation information revealed that the human protein “Q92934” is involved in significant cellular processes including, apoptotic process (GO:0006915), positive regulation of apoptotic process (GO:0043065), mitochondrion (GO:0005739), BAD-BCL-2 complex (GO:0097138), etc. and molecular pathways such as apoptosis (map04210), PI3K-Akt signaling pathway (map04151), chemokine signaling pathway (map04062), and many others. This protein was also found to be localized in mitochondria. Furthermore, the protein “Q92934” was identified to be a pro-apoptotic member of Bcl-2 protein family (InterPro ID: IPR018868). Various studies have reported the regulation of apoptosis in mitochondria, which is tightly controlled by Bcl-2 protein family and its sub-domains such as Bcl-2 homology domain 3 (BH3) (Barber, 2001; Taylor et al., 2006). This results in programmed cell death of infected cells, thus interrupting viral DNA replication and subsequent release of progeny virus (Thomson, 2001). The process of apoptosis is also reported to be significantly regulated by PI3K-Akt signaling pathway and chemokine signaling pathway that are crucially involved in immune responses during viral infection (Vlahakis et al., 2002; Diehl and Schaal, 2013). The drug target for this human protein was found to be “Navitoclax” (DrugBank ID: DB12340), which is reported to be a high affinity inhibitor of Bcl-2 (Wilson et al., 2010). The ChEMBL ID associated with this protein was “CHEMBL5169266” (BCL2/BAD), which is a “protein–protein interaction” type molecule. Further analysis using ChEMBL database revealed that this human protein also interacts with a vaccinia virus protein (UniProt ID: P68451), which belongs to same genus (Orthopoxvirus) as that of monkeypox virus (Manenti et al., 2023).

The interacting viral protein “AAY97628.1” was involved in biological processes such as suppression by virus of host apoptotic process (GO:0019050), and regulation of apoptotic process (GO:0042981). This protein was also found to contain a VAC_F1L type conserved protein and apoptosis regulator M11L like protein. A study in the past showed that the proteins M11L (myxoma virus protein) and F1L (vaccinia virus protein) are localized in mitochondria and play a role in inhibiting host mitochondrial apoptosis by mimicking the function of Bcl-2 protein and blocking pro-apoptotic signals (Everett et al., 2000; Stewart et al., 2005; Douglas et al., 2007). The protein annotations of the human and viral proteins in the above-mentioned interaction pair make it a candidate pair for further analysis and experimental validation of the interaction. Additionally, this shows the reliability of the predicted interactions using the interactome tool and annotations of HuPoxNET. We believe many more potential interaction pairs exist that can be exploited to gain in-depth knowledge of the molecular mechanisms active during viral infection.