In Silico Vaccine Design Tools

Abstract

Vaccines are a boon that saves millions of lives every year. They train our immune system to fight infectious pathogens. According to the World Health Organization, vaccines save 2.5 million people every year and protect them from illness by decreasing the rate of infections. Computational approach in drug discovery helps in identifying safe and novel vaccines. In silico analysis saves time, cost, and labor for developing the vaccine and drugs. Today\'s computational tools are so accurate and robust that many have entered clinical trials directly. The chapter gives insights into the various tools and databases available for computational designing of novel vaccines.

1. Introduction

The vast genome information obtained after the sequencing projects has paved the way to several in silico screening and computational analysis [1]. Nowadays, the vaccine design approaches are based on computational analysis as the methods are time and cost-effective [2].

Bioinformatics is a field that uses information technology and mathematical elements to manage, analyze, and use biological data It is a field that is constantly evolving and producing useful tools for biological sciences [3]. The development and implementation of computational algorithms and software tools aid in the understanding of biological processes, with the primary goal of serving the agriculture and pharmaceutical industries, health care, forensic analysis, crop improvement, food analysis, drug discovery, and biodiversity management [4].

There are various in silico methods for studying linear B-cell epitopes, helper T lymphocytes (HTL), and cytotoxic T lymphocytes (CTL) epitopes [5]. Antigenicity, human population coverage, physicochemical properties, toxicity, allergenicity, and secondary structure of the designed vaccine are all evaluated using cutting edge bioinformatic approaches, to ensure that the designed vaccine is of high quality [6]. In silico tools are also available for prediction, refinement, and validation of the three-dimensional (3D) structures of the designed vaccine candidates [7].

2. In silico vaccine tools

VaxiJen is the first server to predict protective antigens without using alignment. It was created to allow antigen classification based solely on the physicochemical properties of proteins, rather than sequence alignment [8]. Vaxijen is a new alignment-free antigen prediction method based on auto-cross covariance (ACC) transformation of protein sequences into uniform vectors of major amino acid properties. Datasets were generated for viral, bacterial, and tumor proteins. There are validation models to evaluate the results [9].

The server can be used alone or in tandem with alignment-based prediction methods. It is freely available online at the following address: http://www.jenner.ac.uk/vaxijen.

VaxiJen is currently the only tool that can classify protein sequences exclusively based on their physicochemical properties, without any functional or biological information. It is a very fast and easy-to-use tool. The user’s unable to alter the training dataset from which the prediction model is created remains the disadvantage of the tool [10].

2.1 ANTIGENpro

It uses two-stage architecture, multiple representations of the primary sequence, and five machine learning algorithms to predict the antigenicity of proteins [11]. ANTIGENpro is a sequence-based, alignment-free, and pathogen-independent predictor. An Support Vector Machine (SVM) classifier summarizes the results of the analysis and predicts if a protein is antigenic or not, as well as the corresponding probability [12]. The protein antigenicity predictor ANTIGENpro is the first to use reactivity data obtained via protein microarray analysis for five pathogens to determine whole protein antigenicity [11].

2.2 AllergenFP

The AllergenFP is developed based on the dataset that is described by five E-descriptors and the strings that were turned into uniform vectors through auto-cross covariance transformation [12].

Based on the statical analysis of sensitivity and specificity, the overall accuracy confirms that AllergenFP and AllerTOP are the best allergen prediction tools for sequencing compared to the other analysis tools and servers [13].

2.3 AllerTOP

A protein sequence is transformed by auto cross covariance (ACC) into uniform equal length vectors, a protein sequence mining technique developed by Wold et al., 1993.

This technique was used to study quantitative structure–activity relationships (QSARs) of peptides of different lengths. The main characteristics of amino acids were expressed as five E descriptors. The data reveal the hydrophobicity of amino acids, molecular size, helix-forming tendency, the relative abundance of amino acids, and ability to form β-strands. A k-nearest neighbor algorithm (kNN, k = 1) is used to classify proteins based on a training set containing 2427 allergens from different species and 2427 non-allergens [14].

AllerTOP v.2 is a handy, robust, and highly complementary allergen prediction tool with the highest level of precision.

2.4 T- and B-cell epitope identification

The essential for epitope-based antibodies is the accessibility of epitopes. The idea of epitopes present in an antigen should be perceived for such immunization plan. There is a contrast between the acknowledgment of epitopes by B and T cells [15]. B-cell receptors can tie to epitopes in antigen present either in dissolvable structure or on the outside of microbe, and there is no necessity of intercession by some other particle for this limiting [16]. Notwithstanding, the limiting instrument for T-cell epitopes is extraordinary, as they require an epitope to be introduced by MHC particles for restricting to the T-cell receptor [17].

B-cell epitopes are situated on the local protein and are both consistent and conformational. The consistent epitopes are otherwise called straight, or consecutive epitopes involve amino acids present successively in the protein [18]. B-cell epitopes are for the most part surface available, hydrophilic, polar locales of the antigens that can promptly tie to the individual counteracting agent particle [19].

Dissimilar to B-cell epitopes that can be perceived directly, T-cell epitopes require a show of epitope with MHC atoms. Lymphocyte epitopes are just direct or consecutive and the antigens need to go through handling prior to being perceived by their receptors [20]. The protein is the first to cut into little peptides; these peptides tie to MHC particles and hence structure a trimolecular complex with T-cell receptors. There are two kinds Tc cells or cytotoxic T cells that show CD8 protein particles on their surface and Th cells and T-helper cells showing CD4 surface protein. The epitopes that are introduced to Tc cells are shown by Class I MHC particles, while Th-cell epitopes are shown by Class II MHC atoms. The pathways of preparing and introducing epitopes to the two sorts of T cells are unique [21].

2.5 IEDB

The Immune Epitope Database and Analysis Resource (IEDB) is an unreservedly accessible asset that contains a broad assortment of tentatively estimated invulnerable epitopes and a setup of apparatuses for anticipating and dissecting epitopes [22]. The IEDB incorporates counteracting agent and T-cell epitopes for irresistible sicknesses, allergens, and immune system illnesses, and relocate to alloantigen concentrated in people, nonhuman primates, mice, and other species. Life science specialists can utilize the IEDB to foster new antibodies, diagnostics, and therapeutics. The dataset is populated utilizing data caught or curated from peer-reviewed and from information put together by scientists. As of December 2016, more than 18,000 references have been curated, and the dataset contains more than 260,000 epitopes and more than 1,200,000 B cell, T cell, MHC restricting, and MHC ligand elution tests [23].

A comprehensive list of freely available tools for determining the binding affinity of peptides in a protein to different MHC molecules is given in the following Table 1. These tools use machine learning methods such as hidden Markov models (HMMs), support vector machines (SVMs), position-specific scoring matrices (PSSMs), and artificial neural networks (ANNs).

Name of the tool	Description of tool
NetMHC for MHC-I	Prediction of peptide–MHC class I binding using artificial neural networks (ANNs). ANNs have been trained to recognize 81 different Human MHC alleles, including HLA-A, B, C, and E. Predictions for 41 animal alleles (Monkey, Cattle, Pig, and Mouse) are also available [24].
NetMHCPan for MHC-I	NetMHCpan 4.1 is the latest version of NetMHCpan. Using artificial neural networks, the NetMHCpan-4.1 server predicts peptide binding to any MHC molecule of known sequence based on ANNs. The method is trained on over 850,000 quantitative Binding Affinity (BA) and Mass-Spectrometry Eluted Ligands (EL) peptides. BA data includes 201 MHC molecules from humans (HLA-A, B, C, and E), mice (H-2), cattle (BoLA), and primates (P). By uploading a full-length MHC protein sequence, the user can obtain predictions for any custom MHC class I molecule. For peptides of any length, predictions can be made [25].
SYFPEITHI for both MHC I & II	It is constantly updated and comprises a library of MHC class I and class II ligands and peptide motifs from humans and other animals, such as apes, cattle, chicken, and mice. Individual entries are available for all of the motifs that are currently available. MHC alleles, MHC motifs, natural ligands, T-cell epitopes, source proteins/organisms, and references can all be found using this method [26].
ProPred for MHC-II	ProPred is a web-based graphical tool that predicts MHC class II binding regions in antigenic protein sequences. The server uses an amino-acid position coefficient database derived from literature to create a matrix-based prediction algorithm. Predicted binders can be viewed as peaks in the graphical interface or as colored residues in the HTML interface [27].
RANKPEP for MHC-I & II	RANKPEP is an online resource that predicts peptide–MHC class I binding using position specific scoring matrices (PSSMs) or profiles as a basis for CD8 T-cell epitope identification. RANKPEP has been extended to predict peptide-MHCII binding and anticipate CD4 T-cell egress using PSSMs that are structurally consistent with the binding mode of MHC class II ligands [28].
EpiJen for MHC-I	EpiJen is a reliable and consistent multi-step algorithm for T-cell epitope prediction that belongs to the next generation in silico T-cell epitope identification methods. These methods aim to reduce subsequent experimental work by increasing the success rate of epitope prediction [29].
MHCPred for MHC-I & II	A quantitative T-cell epitope prediction server. MHCPred includs alleles from the human leukocyte antigen A (HLA-A) locus. The server currently contains 11 human HLA class I, three human HLA class II, and three mouse class I models. In addition, the new MHCPred includes a binding model for the human transporter associated with antigen processing (TAP). The server also includes a tool for designing heteroclitic peptides. A confidence p value is used to improve the predictability of binding affinities [30].
MULTIPRED2 for MHC-I & II	MULTIPRED2 is a computer programme that predicts peptide binding to numerous alleles of the human leukocyte antigen class I and class II DR molecules with ease. Peptide binding to products of individual HLA alleles, combinations of alleles, or HLA supertypes can be predicted. As prediction engines, NetMHCpan and NetMHCIIpan are used. A1, A2, A3, A24, B7, B8, B27, B44, B58, B62, C1, and C4 are the 13 HLA Class I supertypes. DR1, DR3, DR4, DR6, DR7, DR8, DR9, DR11, DR12, DR13, DR14, DR15, and DR16 are the 13 HLA Class II DR supertypes. MULTIPRED2 predicts peptide binding to 1077 variations representing 26 HLA supertypes in total. It currently calculates population coverage in North America’s five major groups. For the identification of T-cell epitopes, MULTIPRED2 is a useful tool to complement wet-lab experimental approaches [31].
NetMHCII for MHC-II	NetMHCII is an allele-specific approach that uses information from all MHC molecules in the data set. Using artificial neuron networks, the NetMHCII predicts peptide binding to HLA-DR, HLA-DQ, HLA-DP, and mouse MHC class II alleles. For 25 HLA-DR alleles, 20 HLA-DQ, 9 HLA-DP, and 7 mouse H2 class II alleles, predictions can be made. The prediction values are given as a percent -Rank to a set of 1,000,000 random natural peptides in nM IC50 values. The presence of strong and weak binding peptides is indicated [32].
NetMHCIIPan for MHC-II	NetMHCIIpan is a pan-specific method that uses information from all MHC molecules in the data set. Using Artificial Neural Networks (ANN), the NetMHCIIpan-4.0 server predicts peptide binding to any MHC II molecule with a specified sequence (ANNs). It was trained on a large dataset of over 500,000 Binding Affinity (BA) and Eluted Ligand mass spectrometry (EL) measurements, which included the three human MHC class II isotypes HLA-DR, HLA-DQ, and HLA-DP, as well as mouse molecules (H-2). Peptides of any length can be predicted by the network. The model generates a prediction score for the chance of a peptide being delivered naturally by an MHC II receptor of choice. Percent rank score is also included in the output, which normalizes prediction score by comparing it to the prediction of a group of random peptides [33].
MHC2Pred for MHC-II	Promiscuous MHC class II binding peptides are predicted using an SVM-based technique. For 42 alleles, the average accuracy of the SVM-based technique is 80%. Because the dataset was smaller, the method’s performance was lower for a few alleles. The method’s performance was evaluated using 5-fold cross-validation [34].
ClustiMer	Instead of being distributed randomly throughout protein sequences, potential T-cell epitopes typically aggregate in specific immunogenic consensus sequence (ICS) regions as clusters of 9–25 amino acids with 4–40 binding motifs. The ClustiMer algorithm, in conjunction with EpiMatrix, can be used to identify peptides with EpiMatrix immunogenicity cluster scores of +10. These peptides are typically immunogenic [35].
NERVE	Predicts the best vaccine candidates based on a prokaryotic pathogen’s flat file proteome. It is a fully automated reverse vaccinology system designed to predict best vaccine candidates from bacteria proteomes as well as manage and display data through user-friendly output [36].
BlastiMer	One can also use the BlastiMer programme to automatically BLAST “potential epitopes against the human sequence database at GenBank”. BLASTing excludes epitopes with potential autoimmunity and cross-reactivity questions and locates the epitopes that can be used safely in the development of human or animal vaccines BlastiMer can also perform BLAST searches against the PDB, SwissProt, PIR, PRF, and non-redundant GenBank CDS translations [37, 38].
EpiMatrix	EpiVax, an in-silico tool designed to predict and identify the immunogenicity of therapeutic proteins and epitopes. It is also used to redesign proteins and create T-cell vaccines [39].
IEDB Population Coverage analysis	It determines the percentage of people in that location who have shown possible responses to the query epitopes. The Population Coverage Calculation programme is simple and flexible to use. Using MHC binding or T cell restriction data and HLA gene frequencies, a method for calculating anticipated population coverage of a T-cell epitope-based diagnostics are implemented [40].

Table 1.

List of various insilico vaccine design tools.

2.6 BCPred

A continuous B-cell epitope prediction method uses support vector machine (SVM) classifiers that were trained on a homology-reduced dataset of 701 linear B-cell epitopes recovered from the Bcipep database and 701 non-epitopes randomly retrieved from Swiss-Prot sequences using five different kernel approaches and fivefold cross-validation [41].

The advantages of the tool include the samples taken in both the training and test datasets that were experimentally determined. Deep learning methods were implemented for making predictions and allowing the large number of datasets that can improve statistical analysis and features of B-cell epitopes [42].

2.7 ABCpred

The ABCpred server uses an artificial neural network to anticipate linear B-cell epitope areas in an antigen sequence. This server will aid in the identification of epitope regions that can be used to choose synthetic vaccine candidates, diagnose diseases, and conduct allergy research. It is the first server developed based on a recurrent neural network with the fixed length patterns. The training and testing dataset has 700 B-cell and 700 non-B-cell epitopes or random peptides with a maximum length of 20 amino acids. About 65.93% accuracy was achieved using a recurrent neural network [43].

2.8 BepiPred

BepiPred is based on a random forest algorithm that is trained using epitopes from antibody–antigen protein structures. It is a new method based on known 3D structures and the large number of linear epitopes available from the IEDB database hence remains to outperform compared to the other tools. It presents results in a style that is both user friendly and useful for both computer experts and non-experts [44].

2.9 LBtope

It is developed based on the experimentally validated B-cell epitopes and non-B cell epitopes from IEDB. Two types of datasets were derived as LBtope variable with 14,876 and 23,321 B-cell epitopes and non-epitopes of variable lengths, whereas LBtope fixed length has datasets with 12,063 B-cell epitopes and 20,589 non-epitopes of fixed lengths. Further, the very identical epitopes were removed to improve the performance. The tool has accuracy approximately from 54–86% using various features such as dipeptide composition, binary profile, amino acid pair profile [45].

2.10 DiscoTope

The DiscoTope server uses three-dimensional protein structures to anticipate discontinuous B-cell epitopes. Surface accessibility (measured in terms of contact counts) and a unique epitope propensity amino acid score are used in the method. The final scores are computed by adding the propensity scores of nearby residues and the contact numbers [46].

DiscoTope detects 15.5 percent of residues in discontinuous epitopes with a 95 percent specificity. The predictions can guide experimental epitope mapping in both rational vaccine design and the development of diagnostic tools, potentially leading to more efficient epitope identification [47].

2.11 ElliPro

ElliPro predicts linear and discontinuous antibody epitopes based on the 3D structure of a protein antigen. ElliPro accepts protein structures in PDB format as input. If the input is a protein sequence, please go to methods for modeling and docking of antibody and protein 3D structures for more information on these methods. Thornton’s method is implemented as a web platform that allows the prediction and visualization of antibody epitopes in a protein sequence or structure using a residue clustering algorithm, the MODELER program, and the Jmol viewer. ElliPro is based on the geometrical features of protein structure and requires no training. It could be used to predict many forms of protein–protein interactions.

When compared to DiscoTope, which is based on training datasets, ElliPro uses epitope features like residue solvent accessibility, amino acid propensities, inter-molecular contacts and spatial distribution of epitopes, thus improving the prediction ability [48].

2.12 EpiPred

EpiPred is a program that predicts structural epitopes unique to a given antibody. Epitope predictions from EpiPred can be utilized to increase antibody–antigen docking performance. The approach can be utilized using an antibody homology model as input.

Patches on the antigen structure are prioritized based on their likelihood of being the epitope. The program rescores the global docking findings of two rigid-body docking algorithms: ZDOCK and ClusPro, using epitope predictions.

Other approaches, such as DiscoTope or PEPITO annotate broad immunogenic/epitope like regions on the antigen without requiring any antibody information on input unlike epiPred [49].

The use of in silico models can significantly reduce the time and effort required to carry out an epitope discovery experiment. These methods include semi-automatic approaches for epitope discovery and incorporate high-throughput experimental tests for detecting MHC–peptide binding affinities. The in silico methods save a significant amount of resources in terms of both peptide binding classification accuracy and detecting immunogenic peptides. A competent immunologists’ interpretation is required to appropriately confirm the outcome of such a prediction system [50].

3. Future scope

The discovery of vaccines is one of the most important aspects of world public health. Traditional vaccine design procedures have significant shortcomings, but the use of computational tools will overcome these limits. Because immunoinformatics approaches are more useful, modern technologies such as reverse vaccinology, epitope prediction, and structural vaccinology, as well as rational approaches, are in higher demand to produce new vaccine candidates.

The chapter describes the various bioinformatic tools that are available for determining immunogenic characteristics, locating T- and B-cell epitopes, and in silico technologies that are utilized in vaccine development.

4. Conclusion

The application of bioinformatic techniques has greatly accelerated the discovery of new medicinal targets in the post-genomic age. The availability of pathogenic microbe genome sequences has resulted in an increase in the discovery of genes and proteins that could be used to develop drugs or vaccines. Bioinformatic methods have been critical in the analysis of genome and protein sequences to uncover immunogenic proteins among organism’s repertoires. Immunogenicity prediction methods are automated, and the entire proteome may be evaluated to identify top candidates with immunity-inducing features. Not only immunogenic proteins are been identified, but individual epitope mapping has also been completed. Methods for locating T- and B-cell epitopes are now available, which could lead to the development of epitope-based vaccinations.