Recent trends in next generation immunoinformatics harnessed for universal coronavirus vaccine design

ABSTRACT

The ongoing pandemic of coronavirus disease 2019 (COVID-19), caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has globally devastated public health, the economies of many countries and quality of life universally. The recent emergence of immune-escaped variants and scenario of vaccinated individuals being infected has raised the global concerns about the effectiveness of the current available vaccines in transmission control and disease prevention. Given the high rate mutation of SARS-CoV-2, an efficacious vaccine targeting against multiple variants that contains virus-specific epitopes is desperately needed. An immunoinformatics approach is gaining traction in vaccine design and development due to the significant reduction in time and cost of immunogenicity studies and increasing reliability of the generated results. It can underpin the development of novel therapeutic methods and accelerate the design and production of peptide vaccines for infectious diseases. Structural proteins, particularly spike protein (S), along with other proteins have been studied intensively as promising coronavirus vaccine targets. Numbers of promising online immunological databases, tools and web servers have widely been employed for the design and development of next generation COVID-19 vaccines. This review highlights the role of immunoinformatics in identifying immunogenic peptides as potential vaccine targets, involving databases, and prediction and characterization of epitopes which can be harnessed for designing future coronavirus vaccines.

Introduction

History of human coronaviruses and their evolution

In December 2019, a novel contagious coronavirus was first identified in Wuhan, China. This virus (initially known as Wuhan coronavirus) turned out to be the third member of the class of coronaviruses to affect global health following Severe Acute Respiratory Syndrome coronavirus (SARS-CoV) reported in 2002 [1] and Middle East Respiratory Syndrome coronavirus (MERS-CoV) reported in 2012 [2]. The Wuhan coronavirus was later officially designated as Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) given the fact that this virus is genetically related to SARS-CoV.

The receptor-binding domain (RBD) of the spike protein from SARS-CoV-2 is significantly similar to the structure of the spike protein of SARS-CoV [3,4]. However, the spike protein of SARS-CoV-2 possesses a higher affinity to angiotensin converting enzyme 2 (ACE2) than that of SARS-CoV [5]. The disease caused by SARS-CoV-2 was later designated as Coronavirus Disease 2019 (COVID-19) by World Health Organization (WHO). COVID-19 is continuing to spread around the world. As of 23 March 2022, there were more than 474 million confirmed cases and more than 6.1 million deaths across almost 200 countries (mortality rate of over 1%). COVID-19 has shown severity that surpasses previous infections in terms of the propensity to spread readily and cause severe symptoms in the elderly and individuals with morbidities like diabetes and cardiovascular diseases [6,7]. According to standard operating procedures (SOPs) issued by the health authority, treatment concentrates typically on managing symptoms and providing respiratory support [8].

Coronaviruses are a group of enveloped viruses with single-stranded and positive-sense RNA genomes made up of at least six open reading frames (ORFs) [9]. Coronaviruses are well-known for their rapid mutation rate. Since 2019, the genomic sequence of SARS-CoV-2 has mutated repeatedly, giving rise to a significant number of variants reported worldwide [4]. Currently, coronaviruses are only proven to be linked with respiratory diseases in humans. In fact, a total of six human coronaviruses have been described, generally classified into two main groups, pathogenic strain and non-pathogenic strain. In the mid 1960s, two non-pathogenic strains were first reported, termed HCoV-229E [10] and HCoV-OC43 [11]. In November 2002, a respiratory disease first occurred in China followed by a rapid transmission to other countries, ultimately causing more than 8,000 confirmed cases, with a mortality rate of 9.6% [9]. The etiologic agent was recognized as SARS-CoV, a member of betacoronavirus. This was the first report of a coronavirus able to jump the species barrier from bats to humans and cause serious disease, indicating its potential for zoonotic transmission [12,13]. Soon enough, two other non-pathogenic strains, HCoV-NL63 [14] and HCoV-HKU1 [15], were discovered in 2004 and 2005, respectively. A decade later, in 2012, a SARS-like virus emerged and circulated in Saudi Arabia through the manifestation of respiratory disease [16]. Despite the relatively limited transmission among humans, this pathogenic strain, termed MERS-CoV led to two major outbreaks including much later one that happened in South Korea. Overall, the number of cases exceeded 2,000 globally and the mortality rate was as high as 35% [17]. Contrary to both SARS-CoV and SARS-CoV-2 which bind to the human ACE2 receptor, MERS-CoV instead attaches to dipeptidyl-peptidase 4 (DPP4) of the host cells to mediate the infection. Basically, HCoV-NL63 and HCoV-229E were initially detected in bats, whereas HCoV-OC43 and HCoV-HKU1 possibly originated in rodents. The coronaviruses that are phylogenetically related to SARS-CoV and MERS-CoV were found in diverse bat species [18].

Next generation vaccine development based on reverse vaccinology

Immunity plays an essential role in protecting humans against pathogenic infections. The stimulation of protective immunity is usually dependent on immune response against only a small part of foreign proteins in the pathogen, while immunity to the remaining proteins may even induce undesirable allergenic responses or adverse effects to the functions of T-cells and B-cells [19]. With the advances in next-generation sequencing technologies, the amount of genomic data generated that are relevant to immunological research has been substantial. With the aid of a sophisticated immunoinformatics approach, in silico analysis of genomic data can be used to identify potential vaccine targets [20]. This innovative approach can be adopted to design novel subunit vaccines with one or more proteins from the pathogen(s) as required [21–23]. There are several advantages to using immunogenic epitope-based peptides as components of vaccines, including the omission of infectious substances, significantly-reduced risk of cross-reactivity with host tissues, high capacity to perform chemical modification on the products, rapid manufacturing in large scale and relatively long-term storage [24]. In principle, an epitope-based peptide vaccine consists of short antigenic peptides that are able to induce both cellular and humoral immune responses. The length of peptides selected are usually 20 to 30 amino acids from pathogen proteins that carry the highest antigenicity [25–27]. This type of vaccine depends on the major histocompatibility complex (MHC) presentation of the introduced peptides to the CD4 and CD8 T-cells. However, the vast diversity of human MHC loci needs to be taken into consideration in vaccine design to achieve effective immunity in a heterogenous population [28]. Moreover, the efficacy of a vaccine largely relies on the T-cell response to the selected epitopes and targeted antigens. Also needing consideration, is that peptides tend to be biologically active for a short period of time due to enzymatic degradation and that some short peptides may not be recognized by immune cells such as dendritic cells and macrophages, and thus do not elicit considerable immune response [24]. Hence, co-administration of adjuvants is one of the methods used to enhance the immune response [29]. Alternatively, immune response can be intensified when the desired antigenic peptides are linked to larger carrier proteins to improve their stability [30–32]. There is more evidences indicate that multi-component peptide vaccines are able to induce stronger response than a single component peptide vaccine [33].

Immunoinformatics approach

Over the past decades, immunology has been advancing and improving through online resources and software, leading to the establishment of a branch of bioinformatics known as immunoinformatics [34–36]. Immunoinformatics encompasses computational techniques and resources used to study the immune functions. This includes the structural, functional and regulatory analysis of sequence-based or molecular data, modelling of immunologic data and immunologic databases as well as simulation of wet-laboratory experiments [37,38]. Sophisticated tools are employed to depict the immune processes at the molecular, cellular or system level and the development of novel research applications and therapy [36]. In reverse vaccinology, immunoinformatics is utilized in the mapping of potential T-cell and B-cell epitopes via studies and design of predictive algorithms. Antigenic regions and binding sites of the target proteins can thus be identified by analysing pathogen genomes [39,40]. T-cell and B-cell epitopes can also be linked together to form an immunogenic peptide molecule [41]. The immunoinformatics approach is gaining huge interest in various research fields due to the capability in time and cost reduction in immunogenicity studies and increasing reliability in terms of the results generated. Accurate prediction of new epitopes encourages the development of therapeutic methods and accelerate the design and production of peptide vaccines for infectious diseases [42]. Furthermore, the advancement of next-generation sequencing (NGS) has made sequencing of the complete genome of pathogens and humans possible, fast-tracking the development of methods used to predict immunogenic epitopes. The resulting vaccines can be universal and protect against all types of microorganisms as well as provide protection for all individuals regardless of the polymorphism of HLA alleles [27,43]. During the COVID-19 pandemic, many studies related to vaccine design using immunoinformatics approach have also been conducted against other major human pathogens, including Klebsiella pneumoniae, Staphylococcus aureus and Salmonella sp [44–49]., demonstrating once again the reliability and capability of immunoinformatics in designing effective vaccines. In this review, we will describe the vital roles of immunoinformatics approach in identifying immunogenic peptides as potential vaccine targets, involving databases, and prediction and characterization of epitopes. We use keywords such as ‘immunoinformatics’, ‘coronavirus’ and ‘vaccine’ and all seven existing human coronaviruses are included in the scope of this review.

Emergence of SARS-CoV-2 variant of concern (voc) strains

A substitution of aspartic acid with glycine at position 614 (D614G) found in the spike protein of SARS-CoV-2 was firstly discovered in February 2020. Scientists soon discovered that this mutation created a variant which was associated with enhanced infectivity. Within a few months, this variant dominated over the ancestral strain first identified in China, circulating extensively worldwide [50]. To date, five novel variants emerged notably as Variant of Concern (VOC), which are Alpha (UK variant B.1.1.7), Beta (South African variant B.1.351), Gamma (Brazilian variant P.1), Delta (Indian variant B.1.617.2) and Omicron (South African variant B.1.1.529). Needless to say, all of these variants contain D614G substitution in their S protein region. First of all, Alpha variant (B.1.1.7) was initially identified in the United Kingdom. Numerous mutations were observed in the regions including ORF1ab polyprotein, S protein, ORF8 protein and N protein. The S protein had several substitutions as well as deletions at positions 69 and 70 [51]. Substitution P681H is adjacent to the furin cleavage site, which is essential for viral infection [52] while deletion ∆69-70 increases viral infectivity and is associated with immune escape in immunocompromised patients [53]. It is noteworthy that ORF8 protein was truncated due to the presence of a stop codon. According to a preprint [54], Covaxin from Bharat Biotech is effective against Alpha variant.

The Beta variant (B.1.351) contains mutations in ORF1ab polyprotein, S protein, E protein and N protein. This variant was discovered in South Africa with an E484K substitution detected in the spike protein. This particular mutation has exhibited the ability to affect antibody binding [55] and reduce the efficacy of existing vaccines such as Pfizer/BioNTech vaccine, Moderna vaccine, AstraZeneca-Oxford vaccine [56] and Sinovac vaccine [57]. The next variant of concern was the Gamma variant (P.1), initially found in Brazil which has several mutations in ORF1ab polyprotein, S protein and N protein. This variant shares similarities with the B.1.351 variant, such as E484K and K417N/T mutations [58]. These three variants have a N501Y mutation in the spike protein, to be precise, within the RBD. This mutation is shown to alter the conformation of RBD, leading to a higher transmission of viruses. Mutations of K417N/T, E484K, and N501Y in the spike protein result in reduced neutralizing activity by antibodies from convalescent and mRNA vaccine-elicited sera [59]. The Delta variant (B.1.617.2) was first reported in India and it seemed that the mutations are basically different from the three other variants. In addition to ORF1ab protein, S protein and N protein, substitutional mutations took place in M protein, ORF3a protein and ORF7 protein. The mutation of L452R in the S protein reduces the neutralization efficacy by antibodies [60]. While being unique in this variant, T478K is in close proximity to the key mutation of E484K found in other variants such as Beta (B.1.351). Studies have inferred that mutation of P681R showed a certain degree of linkage to more viral loads and increased transmission [61].

Finally, Omicron variant (B.1.1.529) was first reported in South Africa. Mutations are present in ORF1ab protein as well as all structural proteins. In S protein, the number of mutations is strikingly massive, that is 34 mutations, including three deletions and an insertion of three residues at position 214. Unprecedentedly, the RBD also accommodates as many as 15 mutations. For instance, although the glutamic acid at position 484 is substituted by alanine instead of lysine as observed in Beta and Gamma variant, Omicron variant has been reported to exhibit a similar escaping effect from neutralizing antibodies [62]. Q493R and Q498R introduce additional electrostatic interactions and S477N forms hydrogen bonding with ACE2 receptor, increasing the binding affinity towards the receptor [63]. Furthermore, a triple mutation is spotted at the furin cleavage site (H655Y, N679K, and P681H), which may imply an enhancement in the transmissibility [64]. Still, Omicron variant shares some mutations with other VOCs, such as ∆69-70 and N501Y. Table 1 contains all notable mutations observed in these variants [65].

Table 1.

List of non-synonymous mutations in the Variants of Concern (VOCs) (B.1.1.7, B.1.351, P.1, B.1.617.2 and B.1.1.529)

Region	Non-synonymous mutations
	B.1.1.7 (α)	B.1.351 (β)	P.1 (γ)	B.1.617.2 (δ)	B.1.1.529 (ο)
ORF1a polyprotein	T1001I A1708D I2230T ∆3675-3677	Q57H T256I K1655N K3353R ∆3675-3677	S253P S1188L K1795Q ∆3675-3677	-	K856R ∆2083 L2084I A2710T T3255I P3395H ∆3674-3676 I3758V
ORF1b polyprotein	P314L	P314L	P314L E1264D	P314L P1000L	P314L I1566V
S protein	∆69-70 ∆144 N501Y A570D D614G P681H T716I S982A D1118H	D80A D215G ∆241-243 K417N E484K N501Y D614G A701V	L18F T20N P26S D138Y R190S K417T E484K N501Y D614G H655Y T1027I V1176F	T19R ∆156-157 R158G L452R T478K D614G P681R D950N	A67V ∆69-70 T95I ∆142-144 Y145D ∆211 L212I ins214EPE G339D S371L S373P S375F K417N N440K G446S S477N T478K E484A Q493R G496S Q498R N501Y Y505H T547K D614G H655Y N679K P681H N764K D796Y N856K Q954H N969K L981F
M protein	-	-	-	I82T	D3G Q19E A63T
E protein	-	P71L	-	-	T9I
N protein	D3L R203K G204R S235F	T205I	P80R R203K G204R	D63G R203M D377Y	P13L ∆31-33 R203K G204R
ORF3a protein	-	Q57H	-	S26L	-
ORF7a protein	-	-	-	V82A T120I	-
ORF8 protein	Q27stop R52I Y73C	-	E92K	-	-

Vaccine targets identified by immunoinformatics analysis

In order to obtain promising vaccine targets, the identification of conserved regions across variant strains is particularly essential. The proteins that are specific to the target pathogen as well as those that are released or upregulated during the interaction with the host are suited to be vaccine candidates [66]. The potential of the non-conservedproteins to be a therapeutic target relies on the functional importance of the genes and respective proteins in the growth and survival of the pathogen [67,68]. Furthermore, in terms of the immunological mechanism, induction of antibodies as effector molecules is usually needed, suggesting that chosen vaccine protein has to be immunogenic enough to stimulate protective B-cell responses. Prior to this, pattern recognition or augmented dendritic cell activation is important for the optimization of T-helper cell response [69,70]. The best vaccination strategies include both neutralizing antibodies and cytotoxic T-cells [69]. Antigens consisting of multiple epitopes are better than a single epitope. In fact, T-cell epitope content is a significant marker to study the overall immunogenic potential [66].

Structural proteins for human coronavirus vaccine development

Structural proteins of coronaviruses have been gaining attention in the process of designing and developing vaccines. Coronaviruses contain four types of structural proteins, including spike glycoprotein (S) forming into bulky peplomers in the viral envelope, membrane glycoprotein (M), internal nucleocapsid protein (N) and transmembrane envelope protein (E). Some species contain another protein with hemagglutination and esterase functions (HE).

S protein binds to one another naturally into a homo-trimer, resembling Class I membrane fusion protein [71]. S protein comprises two subunits. The S1 subunit is subdivided into N-terminal domain (NTD) and C-terminal domain (CTD). The RBD is located in the CTD. On the other hand, S2 subunit contains the basic elements responsible for membrane fusion, that are, fusion peptide (FP), heptad repeats (HR), membrane proximal external region (MPER) and transmembrane domain (TM). Full-length S protein, S1 subunit, RBD, NTD and FP have been investigated as potential targets of vaccines. The RBD is the most studied due to the fact that it attaches straight to the ACE2 receptor on the host cells. RBD immunization induced antibodies are capable of blocking this interaction and thus effectively minimize the infection [72]. To elaborate, the RBD is conserved relative to the S1 subunit, plus many neutralizing epitopes are found within this domain, making it an excellent candidate for inclusion in a vaccine [73]. N protein is located inside the virus, where it surrounds the viral RNA and forms a helical nucleocapsid. This protein, the most abundant protein in coronavirus, is relatively more conserved than S protein [74]. It is highly antigenic and suitable as a marker for diagnostic purposes [72]. Studies have stated that E protein is a noticeable virulence factor [75]. Along with S and E proteins, M proteins are also found on the viral envelope and are the major component needed for initiating viral-budding. M protein is abundant on the viral envelope (surface) and is well conserved among different species [76].

Given its significance in viral tropism and entry, most studies targeted the S protein of coronavirus, either as the sole target or incorporated with other viral proteins (Table 2). The S1 protein binds to receptors (e.g. ACE2) on the host cell membrane followed by the fusion between virions and cell membranes facilitated by the S2 protein. According to Bhatnager et al. [77], fusion proteins determine the infectivity of coronavirus, making them promising therapeutic candidates. Thus, a multiepitope vaccine was developed from conserved parts of S2 domain to strengthen the immune response from both T-cells and B-cells while eliminating the regions that may cause undesirable side effects such as allergy. Furthermore, Jyotisha et al. [78] have worked on RBD in their in silico vaccine development since this region exhibits no significant homology with the human proteome, diminishing the possibility of autoimmunity. For the latter approach, a group of scientists, Rahman et al. [79] investigated on the development of three types of structural proteins namely S, E and M to produce chimeric peptide vaccines. Instead of the entire length of S protein, RBD and NTD were chosen given that these segments possess immunodominant neutralizing domains. Despite mutations on the RBD which could allow new strains to escape from specific antibodies, a combination of both proteins has proven to generate robust neutralizing antibodies and long-term protective immunity in animal models. E and M proteins can further boost the immunity to tackle viral evasion. Next, main protease, nsp12 and nsp13 which are highly conserved in coronaviruses, are selected with S protein in developing multiepitope vaccine [80]. Safavi et al. [81] did research in the design of a peptide vaccine containing immunogenic epitopes from non-structural proteins located in ORF1ab polyprotein and RBD from S protein. The primary aim was to create vaccines capable of stimulating neutralizing antibodies that inhibit the viral binding to the host cells [81].

Table 2.

List of vaccine candidates studied via immunoinformatics relating to coronaviruses.

Vaccine candidate	Articles reviewed
Spike protein	[131–135]
S2 domain	[77]
RBD	[78]
N protein	[136]
E protein	[137]
M protein	[82]
N protein and S protein	[138]
S protein and E protein	[139]
S protein, M protein and E protein	[79,140]
S protein, M protein and N protein	[141]
S protein, E protein and N protein	[142]
All structural proteins	[143,144]
Whole proteome	[145]
Subtractive proteomics	[146]
3CL hydrolase (main protease/Mpro)	[88]
nsp1	[89]
nsp4	[147]
S protein, E protein, N protein, M protein, selected ORFs (ORF3, ORF4a, ORF4b, ORF5), ORF1a, ORF1ab and ORF8b	[148]
E protein, ORF3a protein, N protein, ORF7a protein and M protein	[90]
S protein, main protease, nsp12, nsp13	[80]
N protein, M protein, ORF3a protein and S protein	[149]
S protein and ORF1ab protein	[81]
ORF1ab protein, S protein, M protein and N protein	[150]
N protein, ORF3a protein and M protein	[91]

There are a number of studies that have selected N protein as the target for potential vaccine candidates. In fact, this gene is relatively more conserved and mutationally stable. N proteins of many coronaviruses are highly immunogenic and are abundant during the course of the infection. N protein can also lead to long-lasting immune response without causing any side effect, especially from T-cells. Being the most abundant protein, M protein appears to carry a pivotal role in intracellular budding [82]. High immunogenicity and multifunctional properties of E protein have made itself a promising vaccine target. Significantly, this protein may be directly associated with virulence as indicated by the deletion of this protein led to increased expression of apoptotic markers and elevated virus-induced inflammation in the infected cells [83]. Moreover, as observed from the accumulation of E protein at the ERGIC, this indicates its role in viral assembly and budding [84,85]. Despite minor portion is incorporated eventually into the viral envelope, it has been found that E protein is also participating in virion release [86,87]. On the other hand, many research studies suggest that all structural glycoproteins are considerably involved in viral pathogenesis. A multi-epitope vaccine should thus be designed, to mainly focus on disrupting the life-cycle of coronaviruses.

Non-Structural proteins and accessory proteins for human coronavirus vaccine development

Other proteins chosen for vaccine development includes 3CL hydrolase, nsp1, ORF3a protein and ORF7a protein. 3CL hydrolase is vital for proteolytic maturation of the virus. Short peptides were extracted from this protein as potential epitopes for both CTLs and HTLs in the design of a multiepitope vaccine [88]. The nsp1 may be a major virulence factor for coronaviruses as seen from its functions. It blocks host gene expression by preventing the translation involving 40S ribosomal subunit as well as degrades mRNA of the host cells. In infected cells, the expression of the IFN genes and the host antiviral signalling pathways were impeded [89]. E protein, ORF3a protein, N protein, ORF7a protein and M protein showed a remarkable linkage with the structural integrity and functionality of the virus [90]. ORF3a protein, N protein and M protein are important in the virus replication and function [91]. Figure 1 illustrates the open reading frames (ORFs) in coronavirus genome and a schematic coronavirus structure labelled with structural proteins.

Figure 1.

Open reading frames (ORFs) composed of non-structural proteins, structural proteins and accessory proteins (from left to right) in coronavirus genome and schematic virion structure of coronavirus. (Note: SARS-CoV-2 genome was used as the template to represent the basic genome organization.) Coronaviruses contain four types of structural proteins, including spike glycoprotein (S) forming into bulky peplomers in the viral envelope, membrane glycoprotein (M), internal nucleocapsid protein (N) and transmembrane envelope protein (E). some species contain another protein with hemagglutination and esterase functions (HE). all non-structural proteins are translated from ORF1ab region (nsp1-16) while the mumber of accessory proteins varies among different coronaviruses. in the case of SARS-CoV-2, the accessory proteins include ORF3a protein, ORF6 protein, ORF7a protein, ORF7b protein, ORF8 protein and ORF10 protein.

Genomics, proteomics and immunomics database

Both molecular biology and immunology produce large amounts of data that have to be stored in general-purpose and designated immunological databases. General-purpose biological databases contain annotated entries of biological sequences. These entries typically contain the sequence, a short description, the source organism, a list of structural or functional features and literature references [37]. A list of databases is available to extract raw data such as protein sequences and structural information as stated in Table 3.

Table 3.

List of databases.

Database	Web address	Reference
Allele Frequency Net Database (AFND)	http://www.allelefrequencies.net/	[92]
AutoPeptiDB	http://autopeptidb.furmanlab.cs.huji.ac.il/	[93]
CoronaVIR	https://webs.iiitd.edu.in/raghava/coronavir/index.html	[104]
Coronavirus database (CoVdb)	http://covdb.popgenetics.net/v3/	[105]
CoVIDep	https://covidep.ust.hk/	[106]
COVIEdb	http://biopharm.zju.edu.cn/coviedb/	[107]
CoV3D	https://cov3d.ibbr.umd.edu/	[108]
DBCOVP	http://covp.immt.res.in/	[109]
GenBank	https://www.ncbi.nlm.nih.gov/	[94]
GISAID	https://www.gisaid.org/	[95]
InterPro	https://www.ebi.ac.uk/interpro/entry/InterPro/	[96]
PDB	https://www.rcsb.org/	[99]
PFAM	https://pfam.xfam.org/	[100]
UniProt	https://www.uniprot.org/	[151]
ViPR	https://www.viprbrc.org/	[102]
viruSITE	http://www.virusite.org/	[103]

The Allele Frequency Net Database (AFND) is an open resource storing the frequency data on the polymorphisms of immune-related genes such as human leukocyte antigen (HLA) system, killer-cell immunoglobulin-like receptors (KIR), MHC class I chain-related genes (MIC) and cytokine gene polymorphisms. AFND collects information and data from four main sources, which are, (i) peer-reviewed publications, (ii) analysed populations by International HLA and Immunogenetics Workshops (IHWSs), (iii) individual submissions and (iv) short publication reports (SPR) in Human Immunology Journal. The allele, gene, genotype or haplotype frequencies for the abovementioned loci can be searched in this database [92]. AutoPeptiDB is another database, composed of 103 high-resolution peptide-protein complexes. The binding affinity of short peptides and proteins is calculated based on the data arising from these complexes in this database. The peptide–protein interactions being considered, generally take place in cellular activities such as signal transduction, protein transport, antigen binding and enzyme-substrate inhibition. There are no two protein monomers that share more than 70% sequence identity in this dataset [93].

GenBank serves as a public database of genetic sequences, focusing on the expansion and dissemination of information. The repository relies on the submissions of sequence data from authors and whole-genome shotgun (WGS) as well as high-throughput data from sequencing centres and issued patents from The U.S. Patent and Trademark Office. GenBank is a partner of the International Nucleotide Sequence Database Collaboration (INSDC) along with European Nucleotide Archive (ENA) and Data Bank of Japan (DDBJ) in which data exchange is done on a daily basis so that a systematic collection of sequence information is accessible worldwide. GenBank also collects and stores amino acid sequences from databases like SWISS-PROT, Protein Research Foundation (PRF) and Protein Data Bank (PDB) [94]. GISAID has gained its reputation as a trustworthy means for international sharing of all influenza virus data including genetic sequence and metadata [95]. In response to the COVID-19 pandemic, related data have also been shared via this public domain recently. InterPro is a unified resource resulting from the integration of protein signature databases including PROSITE, PRINTS, ProDom, Pfam, SMART, TIGRFAMs, PIRSF, SUPERFAMILY, Gene3D and PANTHER. The major application is annotation and functional classification of uncharacterized sequences. Based on sequence positions and protein coverage, protein signatures that fall into the same family or functional domain are grouped into single entry with respective annotation and cross-references [96–98].

Protein Data Bank (PDB) contains structural data of biological macromolecules, typically from X-ray crystallography followed by nuclear magnetic resonance (NMR) and theoretical modelling. The primary data generated are 3D coordinates, source organism, sequence, chemical structures of cofactors and prosthetic groups, names of all components found in structure, qualitative structural information and literature citations. Besides, detailed descriptions of experimental methods for structure determination are included. PDB entries are linked to more than 10 other databases and go through multiple times of review and validation in order to ensure that the description on the 3D structure of the macromolecule of PDB entries in the specified experimental state is reliable and accurate [99]. Pfam is a database of protein families and domains built for the analysis of novel genomes, metagenomes and experimental work on certain proteins and systems. Every Pfam family has a seed alignment of a representative set of sequences. A profile hidden Markov models (HMM) is then automatically constructed from the seed alignment and searched against a sequence database known as pfamseq. All sequence regions that fulfil the family-specific curated threshold are aligned to the profile HMM for full alignment. Pfam entries are manually annotated with functional information from the literature [100]. The UniProt Knowledgebase (UniProt) functions as the central database of protein sequences with functional annotations.The UniProt Knowledgebase has two sections: ‘UniProt/TrEMBL’ comprising computationally analysed records and ‘UniProt/Swiss-Prot’ containing manually annotated records resulting from literature retrieval and curator-evaluated computational analysis. Cross references to external resources are possible in UniProt, for example DNA sequence entries in the INSDC databases, 2D PAGE and 3D protein structure databases, protein domain and family characterization databases, post-translational modification (PTM) databases, species-specific data collections, variant databases and disease databases [101]. The Virus Pathogen Database and Analysis Resource (ViPR) is a repository for viruses that are classified into Category A-C priority pathogens. It keeps record of the information for virus families that are pathogenic specifically to humans including Arenaviridae, Bunyaviridae, Caliciviridae, Coronaviridae, Filoviridae, Flaviviridae, Hepeviridae, Herpesviridae, Paramyxoviridae, Picornaviridae, Poxviridae, Reoviridae, Rhabdoviridae and Togaviridae. Basically, ViPR integrates viral data and related information from three sources including public databases such as GenBank, direct submissions from authors and data derived through computational approach [102]. Finally, as the name implies, viruSITE is a genomic resource specifically for all virus taxa, which gathers data from numerous resources. For instance, genome sequences are retrieved from the reference genomes stored in the NCBI RefSeq database. The data are downloaded, processed and deposited into the database using in-house scripts. It is also a versatile genome search engine and integrated suite of tools for visualization and analysis of sequences between viral genomes [103].

It is noteworthy to know that a few other online platforms were established for coronavirus-related research including CoronaVIR, Coronavirus Database (CoVdb), COVIEdb, CoVIDep, CoV3D and DBCOVP. CoronaVIR collects and stores SARS-CoV-2 information consisting of genomics, diagnosis and therapeutics to serve the purpose of suggesting potential candidates for the development of effective diagnostic tools and vaccines [104]. CoVdb also accommodates gene information of coronavirus, but not limited to SARS-CoV-2 only, such as gene function, subcellular localization, topology, protein structure and evolutionary biology [105]. COVIDep analyses and identifies T-cell and B-cell epitopes based on the immunological data of SARS-CoV and genetic data of SARS-CoV-2. The locations of the identified epitopes are indicated on the primary structure of SARS-CoV-2 proteins as well as graphical display of analysed sequences and genetic variation present on SARS-CoV-2 proteins [106]. Similarly, COVIEdb predicts potential T-cell and B-cell epitopes for SARS-CoV, MERS-CoV and SARS-CoV-2 through their protein sequences. Ultimately, pan-coronavirus vaccine and drug development are facilitated for future coronavirus outbreak [107]. CoV3D regularly updates structural datasets of coronaviruses and enables structural viewing of single and multiple complexes, which then can be utilized for analysis, modelling and structural design. This tool helps the users in understanding antibody recognition on S protein, polymorphisms and glycosylation [108]. DBCOVP is a repository of structural proteins from the genomes of betacoronavirus genera (SARS-CoV, MERS-CoV and SARS-CoV-2) with detailed information such as sequence-structural properties, predicted conserved T-cell and B-cell epitopes, and tertiary structure of epitope-HLA binding-complexes [109].

Analysis and characterization of antigenic protein candidates which can prime T-Cells and B-cells

An epitope is the part of an antigen that is recognized by the adaptive immune system. It binds to specific receptors including antibodies, MHC molecules and T-cell receptors [28]. The binding portion of an antibody is termed a paratope. Epitopes can be either continuous or discontinuous. A continuous or linear epitope is a relatively short (usually 5–6) amino acid sequences recognized by the paratope of a corresponding antibody. In contrast, a discontinuous epitope consists of non-adjacent segments of amino acids, not necessarily from one chain, which form a specific 3D structure, which can also be recognized by antibodies. Since discontinuous epitope arises from a specific 3D fold, it is also known as conformational epitope. Notably, epitopes recognized by B-cell epitopes may contain lipids, nucleic acids or carbohydrates, giving resultant antibodies a vast repertoire while T-cell epitopes are usually peptide fragments. The investigation, identification and development of epitopes are crucial in promoting the advancement of diagnostics and therapeutics [110].

Common parameters that are considered when selecting protein vaccine candidates include antigenicity, allergenicity and physicochemical properties. Under this category, altogether 34 different tools can be employed (Table 4). VaxiJen is a server developed for alignment-independent recognition and prediction of protective antigens of bacterial, viral and tumour origin. The server contains models derived by ACC pre-processing of amino acids properties for protein classification and quantitative structure–activity relationships (QSAR) studies of peptides with different lengths [111]. The ExPASy server contains a variety of databases and analysis tools, focusing on proteins and proteomics. On the other hand, ProtParam is an extensively-used tool that is able to compute physicochemical properties from a protein sequence, without requiring any additional information. These properties include the molecular weight, theoretical pI, amino acid composition, atomic composition, extinction coefficient, estimated half-life, instability index, aliphatic index, and grand average of hydropathicity index (GRAVY). The amino acid and atomic compositions are self-explanatory. The query protein can either be specified as a Swiss-Prot/TrEMBL accession number or as a raw sequence [112]. Finally, similar to VaxiJen, ACC pre-processing is applied to sets of known allergens with different origins and routes of exposure in AllerTop. From here, alignment-independent models for allergen recognition are established based on the chemical properties of amino acid sequences. A mirror set of non-allergens is compiled from the same species. Five machine learning methods including discriminant analysis by partial least squares (DA-PLS), logistic regression (LR), decision tree (DT), naïve Bayes (NB) and k nearest neighbours (kNN) are utilized for the differentiation between allergens and non-allergens. The high sensitivity is seen by the ability to identify new allergens which are structurally diverse in comparison to existing allergens [113]. Overall, these tools can be used to identify potential new vaccine candidates.

Table 4.

List of servers for analysis and characterization of protein candidates.

Tool	Web address	Reference	Articles reviewed
AlgPred	http://crdd.osdd.net/raghava/algpred/index.html	[152]	[80,81,88,135,139,144,146,148,149]
AllerCatPro	https://allercatpro.bii.a-star.edu.sg/	[153]	[81,134]
AllergenFP v1.0	http://ddg-pharmfac.net/AllergenFP/	[154]	[77–81,88,131,142,144,146,149]
Aller Hunter	http://tiger.dbs.nus.edu.sg/AllerHunter	[155]	[131,139,145]
AllerTop v2.0	https://www.ddg-pharmfac.net/AllerTOP/	[113]	[78–80,82,88,91,131,133,136,140,142,144,149,156]
ANTIGENpro	http://scratch.proteomics.ics.uci.edu/	[157]	[77–81,133,142,156]
Class I Immunogenicity	http://tools.iedb.org/immunogenicity/	[158]	[78,88,132,136,145,148]
Epitope Conservancy Analysis	http://tools.iedb.org/conservancy/	[159]	[88,136,138,140,141,144,145,148]
MolProbity 4.2	http://molprobity.manchester.ac.uk/	[160]	[81,90,91]
Protein-Sol	https://protein-sol.manchester.ac.uk/	[161]	[78,79]
Protein Digest	http://db.systemsbiology.net:8080/proteomicsToolkit/proteinDigest.html	-	[131,140]
ProtParam	https://web.expasy.org/protparam/	[112]	[77–82,88,90,91,131,133–135,141–144,146,148–150,156]
SOLpro	http://scratch.proteomics.ics.uci.edu/	[157]	[77,133,135,156]
ToxinPred	http://crdd.osdd.net/raghava/toxinpred/	[162]	[78,82,88,90,131,133,136,140,144,146,148,156]
VaxiJen v2.0	http://www.ddg-pharmfac.net/vaxijen/VaxiJen/VaxiJen.html	[111]	[77–82,88,90,91,131–133,135,136,139,140,142,144–146,148,149,156,163]

T-Cell epitope prediction tools

Decades of studies on MHC-binding motifs and experimental characterization of MHC binders and epitopes [114,115] have progressively boosted the establishment of an information base for predicting peptide binding. This vast amount of information ultimately contributes to the emergence of algorithms for predicting MHC-binding peptides. These developed prediction systems aim to accurately assign binding affinity of a query peptide to a specific MHC molecule. Because of high specificity, HLA-ligand (peptide) binding is the limiting step in the antigen processing pathway. In general, the coverage of the DQ and DP loci is lower than that of the DR locus [43]. Furthermore, determining the peptides that bind to a particular MHC molecule is essential in understanding immune responses before proceeding to the design of peptide-based vaccines and immunotherapies [116]. The peptides presented by MHC class I molecules are generally 8 to 11 amino acids long whereas the ones presented by MHC class II molecules are reported to be longer than 30 amino acids [114]. The prediction of MHC class II-binding peptides is more difficult than class I-peptides, owing to the unknown position of the binding core within the longer peptides [43]. Following in silico predictions, the selected peptides then undergo experimental validation as T-cell epitopes [117]. The prediction tools are grouped according to the type of MHC molecules encountered (Table 5; Figure 2).

Table 5.

list of servers for T-Cell epitope prediction.

Tool	Web address	Reference	Articles reviewed
MHC I binding
CTLPred	http://crdd.osdd.net/raghava/ctlpred/	[164]	[77]
MHC-I Binding Predictions	http://tools.iedb.org/mhci/	[119,165–172]	[77,78,88,135–137,139,173]
MHC-NP	http://tools.iedb.org/mhcnp/	[174]	[136,139]
NetCTL v1.2	https://services.healthtech.dtu.dk/service.php?NetCTL-1.2	[118]	[78,80,81,88,132,136,139,141–143,146,156]
NetCTLpan v1.1	https://services.healthtech.dtu.dk/service.php?NetCTLpan-1.1	[175]	[81,82,139]
NETMHC 4.0	https://services.healthtech.dtu.dk/service.php?NetMHC-4.0	[165]	[90]
NetMHCPAN	https://services.healthtech.dtu.dk/service.php?NetMHCpan-4.1	[176]	[81,141,145]
ProPred I	http://crdd.osdd.net/raghava/propred1/	[177]	[131,133,150]
Proteasomal cleavage/TAP transport/MHC class I combined predictor	http://tools.iedb.org/processing/	[178]	[79,138,139,148]
SYFPEITHI	http://www.syfpeithi.de/bin/MHCServer.dll/EpitopePrediction.htm	[115]	[81]
MHC II binding
IFNepitope	http://crdd.osdd.net/raghava/ifnepitope/	[121]	[77–79,81,82,88,133,141,142,144,146,148,156]
IL-4 Pred	http://crdd.osdd.net/raghava/il4pred/	[179]	[156]
IL-10 Pred	http://crdd.osdd.net/raghava/IL-10pred/	[180]	[141,156]
MHC2Pred	http://crdd.osdd.net/raghava/mhc2pred/	[181]	[81]
MHC-II Binding Predictions	http://tools.iedb.org/mhcii/	[120,169,182–185]	[77–79,88,135,137,139,142,146,148,156,163]
NetMHCII 2.3	https://services.healthtech.dtu.dk/service.php?NetMHCII-2.3	[186]	[80]
NetMHCII PAN 3.2	http://www.cbs.dtu.dk/services/NetMHCIIpan-3.2/	[186]	[81,82,90,141,145]
ProPred	http://crdd.osdd.net/raghava/propred/	[187]	[131,133,150]
MHC I and II binding
MHCPred	http://www.ddg-pharmfac.net/mhcpred/MHCPred/	[188]	[81,90]
Rankpep	http://imed.med.ucm.es/Tools/rankpep.html	[189]	[77,89,91]
TepiTool	http://tools.iedb.org/tepitool/	[190]	[81]

Figure 2.

Classification of T-cell epitope prediction tools based on the type of MHC molecules. tools such as MHC-I binding predictions, NetCTL and NetMHCPAN are designed for MHC class I molecules, while IFNepitope, MHC-II binding predictions and NetMHCIIPAN are the examples of prediction tools for MHC class II molecules. meanwhile, there are in silico tools developed for the prediction of T-cell epitopes that can recognized by both MHC class I and II molecules, for example, rankpep and tepitool.

NetCTL is a web-based tool used to predict human MHC Class I CTL epitopes found in proteins. It combines the predictions of proteasomal cleavage, TAP transport efficiency, and MHC class I affinity. In this case, datasets within this tool are expanded to improve the tool performance. The epitopes predicted are restricted to the A26 and B39 supertypes [118]. NetMHCpan uses both peptide and primary HLA sequence as inputs for artificial neural network-driven (ANN-driven) predictions to generate data while considering all HLA specificities. It is pan-specific, which means that it predicts the affinity of interaction of query peptide with human HLA-A or HLA-B molecule. At population level, it provides the broadest possible population coverage in vaccine design. The ability to handle HLA molecule as well as the availability of specific data for a particular HLA haplotype has lead toward individualized immunotherapy and diagnostics [119]. NetMHCIIpan is ANN-trained on quantitative peptide-binding data covering MHC class II molecules. The identification of the binding core is greatly improved using a network alignment procedure namely offset correction. This procedure integrates the predictions of different neural networks to enhance core register identification without affecting binding affinity. The prediction values are interpreted in the form of IC50 binding affinity. Precise identification of the peptide-binding register leads to the characterization of the MHC class II binding pocket, which is crucial for the discovery and design of peptides that are MHC-recognized [120]. IFNepitope is a web server developed using PHP, Perl, HTML and Java scripts. It is useful for the development of subunit vaccines. Basically, this server is composed of three modules, Predict, Design and Scan. The Predict module allows users to screen peptide library for suitable IFN-γ inducing epitopes. The Design module identifies minimum mutations required in a peptide to produce an IFN-γ inducing epitope. All possible peptides with single-residue mutation are generated and IFN-γ inducing epitopes are predicted from these mutated peptides. Finally, the Scan module predicts regions in an antigen that can activate IFN-γ [121].

B-Cell epitope prediction tools

In contrast to T-cell epitopes, prediction of B-cell epitopes is significantly different. Potential B-cell epitopes have to also engage helper T-cell response and are not necessarily continuous in sequence as described earlier. Folding of proteins allows spatial proximity of amino acids that may be far apart in terms of sequence. The prediction of discontinuous B-cell epitopes is therefore more sophisticated and difficult than that of continuous ones mainly due to machine learning-based methods require continuous sequence data. Importantly, approximately 85% of documented B-cell epitopes are continuous in sequence [122], at some point, can thus be predicted by methods similar to those of T-cell epitope prediction. The working hypothesis applied to existing B-cell epitope predictors is that some amino acids have a higher likelihood of being part of a B-cell epitope [43]. Moreover, a limited understanding of the characteristics of B-cell epitopes hinders the performance of prediction methods. A considerable number of predicted epitopes are subsequently found to be false positives or false negatives. Hence, it is essential to retrieve a comprehensive description of the epitope, for example, its sequence composition and structural characteristics [122]. Most of the prediction tools are designed for linear epitopes instead of conformational epitopes (Table 6; Figure 3). The application of other prediction tools is not restricted to either T-cell or B-cell (Table 7).

Table 6.

list of servers for B-Cell epitope prediction.

Tool	Web address	Reference	Articles reviewed
Linear epitope
ABCPred	https://webs.iiitd.edu.in/raghava/abcpred/ABC_submission.html	[191]	[78,79,81,133]
Antibody Epitope Prediction	http://tools.immuneepitope.org/bcell/	[123,192–195]	[78,80,81,88,91,132,136,139,142–145,148]
BCPRED	http://ailab-projects1.ist.psu.edu:8080/bcpred/predict.html	[128,196]	[77,131,146]
BepiPred 2.0	http://www.cbs.dtu.dk/services/BepiPred/	[124]	[78,79,82,131,137,141,173]
Conformational epitope
Discotope v2.0	https://services.healthtech.dtu.dk/service.php?DiscoTope-2.0	[127]	[77,81,82,131,132,148]
ElliPro	http://tools.iedb.org/ellipro/	[126]	[78–81,88,91,132,141,143–145,148]
Antibody-specific epitopes
IgPred	http://crdd.osdd.net/raghava/igpred/	[197]	[141]

Figure 3.

Classification of B-cell epitope prediction tools based on the type of epitopes. most of the prediction tools are designed for linear epitopes instead of conformational epitopes. antibody epitope prediction, ABCPred and BepiPred are some of the prediction tools for linear B-cell epitope. to predict conformational B-cell epitopes, tools such as ElliPro and DiscoTope can be used.

Table 7.

List of servers for epitope prediction.

Tool	Web address	Reference	Articles reviewed
Antigenic peptides prediction tool	http://imed.med.ucm.es/Tools/antigenic.pl	[195]	[135]
CPORT	https://milou.science.uu.nl/services/CPORT/	[198]	[79]
Epitope Cluster Analysis	http://tools.iedb.org/cluster/	[199]	[138]
NetChop‐3.1	https://services.healthtech.dtu.dk/service.php?NetChop-y3.1	[201]	[81,139]

Bepipred combines the prediction of a HMM and the propensity scale by Parker et al. [123]. It has high accuracy for predicting B-cell epitopes and better performance than other methods when tested on the validation data set. This improved version is trained only on epitope data derived from crystal structures, presumably increasing the quality and predictive power in contrast to the prediction tools trained on linear peptides tested for antibody recognition [124]. ElliPro is a web-based tool for the prediction of conformational epitopes from protein sequence or structure. It implements a modified version of Thorntons method [125], MODELLER program and Jmol viewer to run the prediction and visualization of antibody epitopes in protein sequences and structures. ElliPro considers protein structure as an ellipsoid and calculates protrusion indexes for protein residues outside of the ellipsoid [126]. DiscoTope is a structure-based prediction method for residues located in discontinuous B‐cell epitopes [127]. DiscoTope uses a combination of epitope features such as amino acid statistics, residue solvent accessibility, spatial distribution, intermolecular contacts and surface exposure. It is trained on a compiled data set of discontinuous epitopes from 76 X‐ray protein structures of antibody-antigen complexes. Antibody binding studies and site‐directed mutagenesis are then applied to categorize predicted epitope residues and validate binding. Furthermore, analysis of the local conformations of epitope residues is essential because these conformations must be preserved for discontinuous epitopes [128]. IgPred server relates an antigenic amino acid sequence to its potency to generate systemic (IgG), allergic (IgE) and mucosal (IgA) type of antibody response. Amino acid and dipeptide composition are input features. IgA epitopes are labelled as positive set while the rest of the epitopes are negative set. The prediction of a peptide inducing more than one class of antibodies is possible. If a particular peptide has equal score for two models, it will be assigned to both classes given that the score for both classes is more than the threshold [129].

Conclusion

The emergence of immune-escaped variants of the SARS-CoV-2 has raised concern about the effectiveness of current COVID-19 vaccines rolled out worldwide. Notably, incidents about prior infected individuals as well as vaccinated individuals being tested positive with COVID-19 and some of them even being hospitalized are constantly reported, especially during the severe waves attributed to Delta and Omicron. Instead of scaling up the vaccination rate all over the world, there is a pressing need to develop and redesign more protective vaccines for future prevention. In light of this, a variety of immune system-related databases, prediction software and modelling tools, informatics and computational infrastructure for connecting computer modelling and wet-lab experimentation are critically required to convert high-throughput immune data into effective next-generation vaccines. Immunoinformatics is a promising approach to accelerate the design of novel, universal and efficacious next-generation vaccines and their delivery to pre-clinical studies. With the advances of computational technologies, immunoinformatics is no longer a theoretical approach in dealing with the diagnostics and therapeutics purposes. Its cost-effectiveness and increasing reliability have pushed the development of the field of immunology further than ever. The application of immunoinformatics has been expanding to cover nearly all aspects of immunology including genomics and proteomics associated with the immune system. Therefore, it would be necessary to regularly monitor the development and improvement of immunoinformatics tools and related databases in terms of algorithms and technology so that the deviation between actual and estimated data can be reduced. In this case, close collaboration between immunoinformatics experts and experimental immunologists is vital to nourish the advancement of immunoinformatics, at the same time, to improve its effectiveness and accuracy, for efficient data interpretation. It is noteworthy that immunoinformatics is merely the calculation and prediction of results via in silico tools. Databases are established on the basis the data have already been tested through real experiments. The quality of the prediction largely depends on the quality of data and, the sophistication and precision of the algorithms used. Slight errors in the sequence data generated will affect the efficiency of downstream analysis [130].

Funding Statement

This work was supported by the Malaysia Ministry of Higher Education [203.CIPPM.6711968]; Malaysia Ministry of Higher Education [203/PPSP/6171200]; Universiti Sains Malaysia [1001.CIPPM.8011078].

Disclosure statement

No potential conflict of interest was reported by the author(s).