Retrieval and Investigation of Data on SARS-CoV-2 and COVID-19 Using Bioinformatics Approach

47.2.1 Platforms and Tools for SARS-CoV-2 Literature

Scientists are overburdened with the vast array of papers on COVID-19. In the 5 months since the outbreak, more than 23,000 papers on SARS-CoV-2 and COVID-19 have been published in public domains. This repository is proliferating and has doubled in approximately every 20 days within this period (Brainard 2020). Journals and scholarly publishers have provided outstanding support to such efforts for disseminating scientific contributions rapidly by accelerating their editorial assessment, peer review, and publication processes, which has raised concerns about the quality of the resulting publications (Horbach 2020). As a result, the concept of a curated body of literature is integral in dealing with information overload and deriving meaningful patterns from findings across published papers. Currently, several groups have commenced the utilization of state-of-the-art computational approaches, such as artificial intelligence, for curating hubs for easy and effective access or simply filtering papers with specific focus on high-quality publications (Chen et al. 2020; Lorenc et al. 2020; Johns Hopkins Bloomberg School of Public Health 2020; Balakrishnan 2020).

LitCovid is one such prominent hub of curated literature for tracking recent scientific information on COVID-19 (https://www.ncbi.nlm.nih.gov/research/coronavirus/). This open-resource literature hub was developed by researchers under the aegis of the National Center for Biotechnology Information (NCBI). It assists the retrieval of relevant literature using sophisticated search functions and categorized topics, including an overview, disease mechanism, transmission dynamics, treatment, case reports, and epidemic forecasting (Chen et al. 2020). The articles are also categorized based on geographic location with visualization for better access (Chen et al. 2020). Additionally, a platform developed by the Evidence for Policy and Practice Information and Co-ordinating Centre (EPPI-Centre) at the University College London is an alternative resource for accessing the overview and distribution of articles related to COVID-19 (http://eppi.ioe.ac.uk/COVID19_MAP/covid_map_v11.html) (Lorenc et al. 2020). This platform has been built from the Collaborative Approach to Meta-Analysis and Review of Animal Data from Experimental Studies (CAMARADES) infrastructure that had been developed previously, the infrastructure includes systematic review facility, crowdsourcing for screening, annotation and automation techniques, and a web-based portal. This platform provides an updated map of the current evidence on COVID-19 by categorizing publications based on the study type (Lorenc et al. 2020).

The 2019 Novel Coronavirus Research Compendium (NCRC) is an alternative centralized platform available to the public that rapidly curates and reviews emerging scientific evidence on SARS-CoV-2 and COVID-19, with a specific focus on providing accurate and relevant information from original, high-quality, and trending research (https://ncrc.jhsph.edu/) (Johns Hopkins Bloomberg School of Public Health 2020). The demand for reliable and rapidly curated evidence from the public, programs, policies, and researchers amid the fast growth in the body of literature on COVID-19 has driven numerous researchers from Johns Hopkins Schools of Public Health, Johns Hopkins School of Medicine, and other institutions worldwide to develop this platform (Johns Hopkins Bloomberg School of Public Health 2020). This group rapidly evaluates COVID-19-related publications and strictly selects high-quality articles to be categorized under several topics listed in this platform, including diagnostics, modeling, epidemiology, pharmaceutical interventions, non-pharmaceutical interventions, clinical presentation and prognostic risk factors, vaccines, and ecology and spillover (Johns Hopkins Bloomberg School of Public Health 2020).

The WHO initiated an international clinical trial called SOLIDARITY a week after the declaration of the COVID-19 pandemic, and over 100 countries have joined the SOLIDARITY trial and are working simultaneously to solve the issue at hand (Balakrishnan 2020). A clinical trial is a type of research methodology in which tests and treatments for the disease are evaluated by assessing the safety and efficacy of clinical candidate interventions (Friedman et al. 2010). It relies on a foundation of evidence to improve the quality of health care and help stakeholders control the costs involved through careful comparison with alternative interventions (Friedman et al. 2010). Amid the rapid growth in the number of COVID-19 patients globally and the absence of a confirmed drug or treatment method, the urgency to develop or discover an efficient therapeutic strategy for COVID-19 has triggered a spike in clinical trial research and unprecedented growth in the number of findings from studies on a specific disease within a brief period (Thorlund et al. 2020). Hence, an easily accessible platform that summarizes findings from COVID-19 clinical trials is necessary for convenient tracking of relevant information without including irrelevant data in the findings. Several groups of researchers have taken the initiative to develop media that is focused on the mapping and summarization of clinical trial research instead of providing an overview of all articles related to COVID-19. It includes COVID-evidence (covid-evidence.org) and Global Coronavirus COVID-19 Clinical Trial Tracker (covid-trials.org) (Thorlund et al. 2020; COVID-evidence 2020; Ruano et al. 2020). Both platforms use automatic search and expert manual extraction strategies for retrieving data from several sources, including published articles and the International Clinical Trials Registry Platforms, to minimize duplicated entries (Thorlund et al. 2020; COVID-evidence 2020). Additionally, ClinicalTrials.gov (clinicaltrials.gov) and the International Clinical Trials Registry Platform (ICTRP; www.who.int/ictrp/en/) by WHO are relevant resources for tracking clinical trials, although these platforms provide data without recapitulation to avoid unnecessary duplication (Thorlund et al. 2020; World Health Organization 2005; ClinicalTrials.gov 2000).

Additionally, the development of text mining and information retrieval systems is the key to the dynamic discovery and extraction of relevant, nontrivial information from the massive collection of literature. To accommodate such advancements for use in SARS-CoV-2- and COVID-19-related studies, the Allen Institute for AI (AI2) released a COVID-19 Open Research Dataset (CORD-19; www.semanticscholar.org/cord19) in collaboration with governments’ several leading institutions (Wang et al. 2020). This resource contains more than 128,000 scholarly articles on COVID-19, SARS-CoV2, and related historical research on coronavirus, which are assembled in a machine-readable format to provide an accessible structure and system for the retrieval process using computational methods such as data mining, machine learning, and natural language processing (Wang et al. 2020). The development and availability of this resource are provided to the global community to apply recent advances of AI techniques and to develop robust and user-friendly computational tools that help researchers find answers to their questions from published studies. The CORD-19 itself consists of several built-in tools such as SciSight and SPIKE-CORD for exploring the evolving network of scientific information and for effectively searching data, respectively (scisight.apps.allenai.org; spike.covid-19.apps.allenai.org/search/covid19) (SciSight 2020; SPIKE-COVID 2020). SciSight helps retrieve a comprehensive network of information using an AI-powered visualization framework. This tool has several features for exploring researcher working networks (the network of science), searching key facets (Faceted Search), and exploring the association between proteins, genes, and cells (proteins/genes/cells), as well as between diseases and chemicals (Diseases/Chemicals) (SciSight 2020). Conversely, SPIKE-CORD offers user-friendly text mining options for effective search and interaction with CORD-19 data. That is an advanced tool and comprises three query modes, including Boolean queries, sequential queries, and structured queries (SPIKE-COVID 2020).

ViralZone (www.expasy.org/viralzone/) is another important resource that provides a clear view of the biological processes of the complete identified virosphere (Hulo et al. 2011). This web resource provides fact sheets with information on sequence information, replication cycles, taxonomy, and epidemiology, as well as graphics describing virion organization, genome transcription, and translation strategies for every identified virus families/genera, thereby creating an accurate and concise information platform to improve the understanding of the complications associated with the massive diversity in viruses. The ViralZone platform has a dedicated resource for SARS-CoV-2 with additional information on antiviral drug development and COVID-19 treatment strategies (Hulo et al. 2011).

To deal with the devastating effects of SARS-CoV-2 infection, the global scientific community has made an effort to contribute to the current information on the disease, which has created a massive collection of readily accessible literature. A challenge involving the tracking of novel findings and the construction of a meaningful network from such a massive data repository awaits clinicians, researchers, and policymakers. Fortunately, several research groups are addressing the challenge and have attempted to create sophisticated platforms and tools that focus on providing user-friendly media. Given these efforts, the utilization of such media aids active tracking of important findings on COVID-19.

47.2.2 Sequence and Structure Database of SARS-CoV-2

The ability to sequence DNA at higher throughput and lower costs leads to the rapid increase in the available sequence data. Sequence analysis plays a vital role in viral surveillance, host reservoir identification, and public health policy debates (Brister et al. 2015). The development of a sequence database is essential in current bioinformatics research and applications for storage and retrieval locations of sequence data and annotation. Generally, three primary public nucleotide sequence databases facilitate the availability of DNA sequence data to the public: NCBI, the European Bioinformatics Institute (EBI), and the DNA Data Bank of Japan (DDBJ). Over the years, the repository of sequences deposited in these databases has grown at an increasing rate. Since these databases form the International Nucleotide Sequence Database Collaboration and exchange updates daily, they follow certain common principles for the arrangement of sequence data and mainly provide the same accession number (Koonin and Galperin 2003).

Additionally, scientific journals generally specify that the sequence data should be cited with the accession number of information from these databases in a paper that describes a nucleotide or protein sequence (Brister et al. 2015). Following the growth in the number of viral genome sequences and the challenges of implementing a purely well-annotated representation of viral genome sequences, NCBI has a unique feature known as the NCBI Viral Genome Resource (www.ncbi.nlm.nih.gov/genome/viruses/) to serve better the needs of the community of virologists. This resource has a central browser for viral and viroid genomes, which lists all viral and viroid species indicated by the corresponding reference sequence and includes links to genome neighbor sequences (Brister et al. 2015).

Specific databases for viral sequences have also been established; this is an alternative of the general public nucleotide sequence databases. The development of a specific database for viruses is essential to avoid the challenges arising from the sharing of viral sequence information. For example, the Global Initiative on Sharing All Influenza Data (GISAID) EpiFlu (https://www.gisaid.org/) was initially developed to address problems experienced while sharing virus sequences, such as the potential intervention by government agencies for the international exchange of information and the hesitation of researchers to share data on sequences of lethal viruses (Elbe and Buckland-Merrett 2017). This database provides protection and assurances to data contributors through a unique data access agreement. Users must register themselves with official affiliation to access the data in this database and comply with several agreements, including the acknowledgment of the contributor while using their data in publications and refraining from data sharing with third parties outside the GISAID community (Elbe and Buckland-Merrett 2017). This mechanism allows researchers and the government to share their sequence data promptly and is important because the immediate availability of viral genomic data helps expedite the processes involved in tracking and recognizing emergent epidemics or pandemics (Elbe and Buckland-Merrett 2017). The GISAID EpiFlu system itself contains above 27,000 partial and complete genomic sequences of SARS-CoV-2 that have been contributed by clinicians and researchers worldwide, as of mid-May 2020.

Given the extent of available sequence information, the number of structurally characterized proteins, nucleic acids, and other biological macromolecules is relatively low. Acquiring three-dimensional molecular structure information helps characterize the molecular function and interaction. Additionally, elucidating the viral protein structure and interaction is fundamental to comprehending the virus receptor recognition mechanism, which is associated with its infectivity, pathogenesis, and host range, and is a requisite in the effort to develop structure-based drugs. The Protein Data Bank (PDB) is a prominent global repository of experimentally determined three-dimensional structures (Sussman et al. 1998). It is one of the first open-access digital resources for biological sciences established in 1971. The Worldwide Protein Data Bank currently manages this database (wwPDB; http://www.wwpdb.org/), which includes the RCSB Protein Data Bank (RCSB PDB; https://www.rcsb.org/), the Protein Data Bank Japan (PDBj; https://pdbj.org/), the Protein Data Bank in Europe (PDBe; https://www.ebi.ac.uk/pdbe/), and the Biological Magnetic Resonance Data Bank (BMRB; http://www.bmrb.wisc.edu/) (Burley et al. 2017). This community manages annotated protein structure information and provides convenient access to experimental data to the community of researchers and students of biological sciences. In response to the COVID-19 pandemic, RCSB PDB has developed COVID-19/SARS-CoV-2 Resources, which provide quick access to all PDB structures of SARS-CoV-2.

47.3.1 Comparative Genomics of SARS-CoV-2

Immediately after the initial report on 27 unusual cases of pneumonia in Wuhan on December 31, 2019, the complete genome sequence of the causative agent, SARS-CoV-2, was made available on January 10, 2020 (Wu et al. 2020; Lu et al. 2020). Measuring the genetic distance is a simple computational approach for understanding the origin of biodiversity. For this purpose, the detection of candidate homologous sequences is a fundamental step, which is achieved by searching sequence similarity in sequence databases or GenBank (Pearson 2013). Several available programs statistically estimate sequence similarity against existing sequences in databases, such as Basic Local Alignment Search Tool (BLAST), SSEARCH, FASTA, and HMMER (Altschul et al. 1997; Smith and Waterman 1981; Pearson and Lipman 1988; Johnson et al. 2010). Sequence candidates with significant similarity can be obtained using these programs, as the sequence data in GenBank is becoming increasingly comprehensive, based on the principle that scientific journals that require the data of sequences mentioned in the article should be registered with this platform (Pearson 2013). Once the homologs are detected, building a more accurate alignment model using multiple sequence alignments forms the basis for the calculation of genetic distance and determining genome organization, as well as developing other evolutionary analyses, such as the construction of phylogenetic trees and recombination detection. As computation has become more accessible recently, utilizing a more rigorous multiple sequence alignment method that employs iterative approaches such as MAFFT and MUSCLE is recommended for acquiring high throughput and high accuracy results (Pearson 2013; Katoh et al. 2002; Edgar 2004). Through these steps, the pairwise sequence identities of SARS-CoV-2 initially isolated from several patients were analyzed and were observed to share a nearly identical similarity of above 99.9%, and this indicates a recent host shift into the human population (Lu et al. 2020). Several groups have confirmed that SARS-CoV-2 belongs to the genus Betacoronavirus, with the bat SARS-related coronavirus (SARSr-CoV RaTG13) as the most closely related member (96.2% similarity). Human-infecting coronaviruses that are closely related to this virus are SARS-CoV (approximately 79% similarity) and MERS-CoV (approximately 50% similarity) (Lu et al. 2020; Fahmi et al. 2020; Zhou et al. 2020; Paraskevis et al. 2020; Chan et al. 2020).

As SARS-CoV-2 was observed to be related to the genus Betacoronavirus and the genome organization of betacoronaviruses has been elucidated previously, the genome organization of SARS-CoV-2 can be determined simply by sequence alignment with the genomes of other members of Betacoronavirus (Kim et al. 2020; Marra et al. 2003; Rota et al. 2003). It was performed by alignment to two representative members of Betacoronavirus, SARS-CoV Tor2 (GenBank accession number AY274119), and bat SL-CoVZC45 (GenBank accession number MG772933), which are associated with humans and bats, respectively (Wu et al. 2020). The SARS-CoV-2 genome contains a positive-sense single-stranded RNA with a 5′ cap structure and a poly-A 3′ tail, and it has ~29,900 nt encoding ~9860 amino acids (Marra et al. 2003; Rota et al. 2003). The 5′ end of the SARS-CoV-2 genome contains a predicted RNA leader sequence of ~70 nt that is fused with two open reading frames (ORF1a and ORF1b) through short motif transcription regulatory sequences (TRSs) (Wu et al. 2020). The TRSs precede each structural or accessory gene, which aids gene expression (Fehr and Perlman 2015). Usually, the ORF1a and ORF1b overlap in betacoronaviruses, and the region occupies two-thirds of the genome. This region encodes the viral replicase and a translational read-through comprising a − 1 ribosomal frameshift, which allows the translation of this overlapping reading frame into a single polyprotein (Fehr and Perlman 2015; Thiel et al. 2003). Subsequently, upon infection of an appropriate host cell and translation, this large polyprotein commonly undergoes proteolytic processing and is cleaved by virus-encoded proteases into several nonstructural proteins (nsps), including viral papain-like protease (PLpro), main protease (3CLpro, also known as 3-chymotrypsin-like protease), RNA-dependent RNA polymerase (RdRp), and helicase (Hel) (Marra et al. 2003). These proteins mediate the replication process of the viral genome and generate nested transcripts, which are essential for the synthesis of viral proteins (Marra et al. 2003). There are structural and accessory genes downstream of the ORF1a and ORF1b sequences that encode the spike glycoprotein (S), ORF3a, ORF3b, envelope (E), membrane (M), ORF6, ORF7a, ORF7b, ORF8a, ORF8b, and nucleocapsid (N), which are interspersed with the TRS motifs (Fehr and Perlman 2015; Thiel et al. 2003). The structural proteins are responsible for packaging the viral particles and the entry process in host cells, while the accessory proteins are likely to play vital roles in viral pathogenesis (Marra et al. 2003).

A notable feature within the specific region of SARS-CoV-2 can be identified by comparative analysis once the genome organization is known. The S protein is one of the most variable regions within the genomic sequence of SARS-CoV-2. It displays the lowest sequence identity of around 75% to bat-SL-CoVZC45 and bat-SL-CoVZXC21 (Lu et al. 2020). The trimeric S proteins of CoVs mediate binding to the cell receptor ACE2 and membrane fusion in the viral entry process; understanding this protein is compulsory in vaccine design as it elicits an antibody response (Li et al. 2005a; Belouzard et al. 2012; Babcock et al. 2004). The core structure and receptor-binding motif of S protein to the claw-like structure of ACE2 receptors occur in the receptor-binding domain (RBD) (Li 2015, 2008). There are six vital amino acids in the RBD for enhancing the viral binding of SARS-CoV to human ACE2 receptors. They are Y442, L472, N479, D480, T487, and Y491 based on S protein sequence (Wan et al. 2020). With coordinates based on SARS-CoV, those amino acids are L455, F486, Q493, S494, N501, and Y505 in SARS-CoV-2, of which is only one amino acid identical to SARS-CoV (Wan et al. 2020; Andersen et al. 2020). Another notable feature in the S protein of SARS-CoV-2 that has been identified is the presence of polybasic cleavage site (RRAR) at the junction of the two subunits of the S protein: S1 and S2 (Walls et al. 2020). This cleavage site is preceded by inserted proline, which is predicted to result in the o-linked glycosylation of S673, T678, and S686 in the S protein sequence (Andersen et al. 2020). Furin proteases cleave this site during biosynthesis; this process is essential to enable exposure of the fusion sequence for cell membranes, which crucially mediates the cell entry process (Walls et al. 2020). The fact that the cleaving process varies from coronavirus to coronavirus implies that the cleavage site is associated with transmissibility and pathogenesis in the host animal.

CoVs have developed several genetic mechanisms to control replication errors as they have an extraordinarily large RNA genome. Recombination is the capacity to create chimeric molecules from two parental genomes of different origins during coinfection and is one such mechanism that contributes to the genetic stability and diversity of CoVs (Simon-Loriere and Holmes 2011; Lai et al. 1994). Deciphering recombination event that contributes to the virus emergence hence is one way to identify SARS-CoV-2 origins. This analysis is also important to identify virus expansion of viral host range and the evolution of resistance to antivirals (Brown 1997; Gibbs and Weiller 1999; Nora et al. 2007). Numerous tools for examining recombination have been developed within the last two decades, such as SIMPLOT, RDP, TOPALi, and 3seq, all of which are optimized to detect recombination in different ways (Lole et al. 1999; Milne et al. 2008; Martin and Rybicki 2000; Lam et al. 2017). The comprehensive list of tools for analyzing recombination events can be found at http://bioinf.man.ac.uk/robertson/recombination/programs.shtml. Even though CoVs have been reported to be a highly recombinogenic group of viruses, detecting recombination event of the recently emerging CoV from the ancestor that likely develops multiple recombination events, however, is not trivial (Zhang and Holmes 2020; Forni et al. 2017; Hon et al. 2008). It has been reported that SARS-CoV-2 is not a recombinant of any viruses detected to date (Tang et al. 2020). This indicates that the sampled diversity of CoVs, especially within the subgenus sarbecovirus, is massively inefficient.

47.3.2 Molecular Phylogenetics

Elucidating the origin, relationships, and transmission routes of emerging infectious agents is a key to conveniently understanding their biological processes and the potential methods of intervention. These functions are related to molecular evolution, which can be elegantly illustrated by molecular phylogenetic analysis (Kühnert et al. 2011; Lemey et al. 2009). The phylogenetic tree has played a fundamental role in the efforts of inferring the evolutionary history of several infectious agents such as HIV, HCV, and SARS-CoV (Kühnert et al. 2011; Gao et al. 1999; Santiago et al. 2002; Pybus et al. 2009; Markov et al. 2009; Li et al. 2005b). The phylogenetic methods are pertinent for detecting orthology and paralogy, estimating divergence times, reconstructing ancient proteins, finding the important residues to natural selection, identifying recombination points, determining the identity on new pathogens, and identifying mutations likely to be associated with disease (Holder and Lewis 2003). Once the predetermined sequences are aligned, a wide range of methods and software packages are available for reconstructing the phylogenetic tree, which can, at times, make it challenging for researchers to choose the well-suited ones. Each method relies on a different algorithm and has its own advantages and limitations. There are two types of reconstruction methods: distance-based methods and character-based methods (Lemey et al. 2009). Distance-based methods reconstruct trees from the calculated distance matrix of every pair of sequences. These techniques can also be classified into two categories: clustering-based (UPGMA and neighbor-joining) and optimality-based (Fitch-Margoliash and minimum evolution) algorithms (Lemey et al. 2009). These methods generally have a rapid calculation time and a large number of applicable substitution models for calculating the genetic distance scores; however, their applicability in cases with divergent sequences and high variance of large distance estimates is poor (Lemey et al. 2009; Yang and Rannala 2012). Conversely, character-based methods are based on the mutational event scoring of given aligned characters (sites in alignment), which prevents information loss in pairwise distance calculation (Yang and Rannala 2012; Xiong 2006). Character-based methods include maximum likelihood (ML), maximum parsimony (MP), and Bayesian inference (BI). MP follows the “Ockham’s razor” principle by selecting the proposed tree with a minimum score of discrete changes to the given alignment data (Lemey et al. 2009). The resulting tree in MP generally has the minimum instances of homoplasy (Lemey et al. 2009). Owing to its simplicity, the computations in MP are less extensive than those in other character-based methods, although these lack explicit assumptions. The ML method estimates trees with the highest likelihood of development using discrete change data of alignment based on multiple proposed trees depicting different evolutionary hypotheses (topologies, branch lengths, and sequence substitution models) of taxa in an alignment. Conversely, BI calculates the posterior probability distribution of the proposed trees based on the model, prior probability distribution, and data. When the data are informative, the majority of the posterior probability is typically concentrated in one tree (or a small subset of trees in an ample tree space) (Lemey et al. 2009; Xiong 2006). At present, ML and BI are considered to represent the best accurate methods since they incorporate the most complex statistical models.

Several authors have constructed the phylogenetic trees of SARS-CoV-2 with that of previous CoVs using the whole genome and specific region sequences to understand its evolutionary history and recombination event (Li et al. 2020b; Wu et al. 2020; Lu et al. 2020; Tang et al. 2020). Based on the currently available genome samples of CoVs, the whole-genome phylogenetic tree indicates that SARS-CoV-2 is closest to SARSr-CoV RaTG13, followed by Pangolin SARSr-CoVs. Two possible scenarios can plausibly explain the origin of SARS-CoV2: natural selection in an animal host prior to zoonotic transfer and natural selection in humans following zoonotic transfer (Andersen et al. 2020). With the first scenario, the genuine progenitor of SARS-CoV2, however, is unrevealed; this uncertainty is attributed to the absence of animal coronaviruses that have the highest similarity on the entire genome location to SARS-CoV-2; this also indicates that subgenus sarbecovirus is likely to have massive hidden diversity (Boni et al. 2020; Andersen et al. 2020). Even though the RaTG13 bat coronavirus has been identified as the virus with the highest similarity to SARS-CoV-2, the RBD in the S protein of SARS-CoV2 is significantly related to that in the Pangolin (Manis javanica) coronavirus, which belongs to the sister lineage of the RaTG13 bat coronavirus (Wan et al. 2020; Andersen et al. 2020; Zhang et al. 2020). RBD is a region critical in SARS-CoV-2 transmission in humans because it has a high affinity for human ACE2 (Wan et al. 2020). This binding is likely a result of natural selection of human or human-like ACE2, as a novel binding pattern distinct from those previously predicted was observed (Wan et al. 2020; Andersen et al. 2020). This complication highlights the uncertainty involved in the identification of the direct ancestor of SARS-CoV-2. Conversely, pangolin SARSr-CoVs are likely to be the progenitor of SARS-CoV-2 if the second scenario is applied, following to their significant similarities of the RBD region (Andersen et al. 2020). Since pangolin SARSr-CoVs have RBD that binds with high affinity to humans ACE2, it is likely that this virus jumped to the human population and acquired genomic features that give rise to SARS-CoV-2 (Andersen et al. 2020).