Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/319169721

A systematic review: B-cell conformational epitope prediction from epitope


characteristics view

Conference Paper · July 2017


DOI: 10.1109/ICSTC.2017.8011859

CITATIONS READS

2 126

5 authors, including:

Binti Solihah Afiahayati Afiahayati


Universitas Gadjah Mada Universitas Gadjah Mada
6 PUBLICATIONS   2 CITATIONS    19 PUBLICATIONS   95 CITATIONS   

SEE PROFILE SEE PROFILE

Sri Hartati
Universitas Gadjah Mada
120 PUBLICATIONS   324 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

thinning of various local character View project

MetaVelvet-SL View project

All content following this page was uploaded by Binti Solihah on 18 February 2018.

The user has requested enhancement of the downloaded file.


A Systematic Review: B-Cell Conformational
Epitope Prediction from Epitope Characteristics View

Binti Solihah1,2, Edi Winarko1, Afiahayati1, Sri Hartati1, Moh. Edi Wibowo1
1
Department of Computer Science and Electronic
Universitas Gadjah Mada,
Yogyakarta, Indonesia
2
Department of Informatics,
Faculty of Industrial Technology, Universitas Trisakti,
Jakarta, Indonesia

{binti.solihah, af1a}@mail.ugm.ac.id, {ewinarko, shartati,mediw}@ugm.ac.id

Abstract—B-cell conformational epitope identification is the This literature review focus on the discussion of the
crucial issue in vaccinology. Limitation on experimental methods relationship between epitope characteristics extracted from the
in the biological side, dataset and availability of computational computational analysis, feature representation and the
resources opens a chance on developing prediction method which accuracy of epitope prediction method. The goal of the review
can accelerate epitope identification. A number of methods have is to identify possibility to improve feature representation and
been developed but their performance is still medium. Epitope to identify problems in the development of epitope prediction
prediction is a knowledge-based method. Presenting the method. To achieve the aim of the review, a systematic
statistical or computational based epitope characteristics together literature review is conducted by the guideline of Kitchenham
with epitope prediction method will facilitate the newcomer on
and Charters[7]. The discussion is started with the related
identifying importance feature, improve the existing feature and
propose the new feature. To reach the goal of the review, the
work and continued with the explanation of methodology. The
research papers are collected from both epitope analysis research next section is the result of review and conclusion is given at
and epitope prediction methods research. The prediction the end of the review.
methods are evaluated on what characteristics of epitope have
been implemented on feature representation and are shown in a II. RELATED WORK
mapping table.
There are two research areas related to this review, i.e.
Keywords— conformational epitope prediction method, epitope research on the computational analysis of epitope and research
characteristics, 3D structure based feature on conformational epitope prediction method. Review on
computational analysis aims to explore the epitope
determinant. Review on conformational epitope prediction
I. INTRODUCTION aims to explore the state of the art of methods.
B-cell epitope identification is needed in the modern
vaccine development. B cell epitope is part of the antigen that The first paper on conformational epitope is work done by
interacts with the paratope of antibody. About 90% of B-cell Hubbard et al.[8]. The next computational analysis conducted
epitope is conformational epitope that is adjacent spatially but by [3, 9,10,11, 12]. The computational analysis on epitope by
located in separated segment of primary sequence[1]. B-cell Rubinstein et al. [3] analyzed the physicochemical properties,
epitope data is extracted from Antigen-Antibody complexes structure, and geometric characteristics of epitope. Fifty three
from X-ray crystallography. The data represents the data were selected from 256-row data on the SPIN server. Sun
coordinate record of its atom components. These coordinate et al. [9] analyze physicochemical properties (AAIndex) and
can not be used directly to identify epitope of the antigen. The structure aspect of epitope characteristics on 161
availability of 3D structure of Antigen-Antibody complex in immunoglobulin complexes. Kringelum et al. [13] analyze the
Protein Data Bank (PDB) supports biology research [2] and structure and geometric characteristic including size, shape,
makes it possible to know epitope characteristics and to reveal and orientation to the antibody on 107 antigen data from 376
new aspects in molecular immunology identification [3]. More Ag-Ab complexes. Sivalingam and Shepherd [10] identify
general description about prediction method on B-cell discontinuity level of the conformational epitope on 150 Ag-
conformational epitope from database point of view, Ab complexes from the Summary of Antibody Crystal
algorithms, web server implementation, and its application on Structures (SACS) database. Zheng et al. [12] analyze the
problem-solving is discussed in [4,5,6]. deep level of the conformational epitope on 161 protein
sequence from 144 Ag-Ab complexes. Epitope characteristics
from computational analysis affects the feature extraction
selected on the proposed prediction methods. TABLE 1. PICOC criteria

The Majority of the epitope prediction method use aspect specification


classification approach to label input residue as epitope or Population (P) computational analysis on epitope and conformational
non-epitope. From the data sample availability, this problem is epitope prediction
Intervention (I) Epitope characteristic in several aspects, method, and
categorized as imbalance class problem, where epitope sample feature extraction
is much less than the non-epitope sample. For example, the Comparison(C) n/a
ratio of the epitope to non-epitope exposed residue on [3] is Outcome(O) Problem, Feature,
1:10. Identification of existing conformational epitope Relation between epitope characteristic with epitope
prediction method shows that the early developing methods prediction performance
ignore imbalance class problem and focus on the developing Context (C) Literature review on computational and statistical
analysis of epitope
of discriminant function on the basis of feature combination.
Literature review on B cell conformational epitope.
Methods that use this approach are DiscoTope [1], Pepito [13],
Discotope 2 [14], Epitome [15], Ellipro [4], EPCES [16],
SEPPA [17], and SEPPA 2[18]. B. Conducting the review
The newest approach uses machine learning approach by The first step in conducting the review determines the
Rubinstein et al. [19] and considers the imbalance class search strategy. The keywords for searching are extracted
problem, i.e. [20,21,22]. Other than feature and imbalance from RQ. The used keywords to collect literature are B-cell
class problem, feature selection is one of the open problems to epitope characteristics, conformational epitope prediction,
build effective classifier on epitope prediction. Another and survey on epitope prediction method. After the keywords
approach that has been applied to epitope prediction method is are determined, the database of literature is constructed. The
a graph-based method, as proposed in[23,24,25]. Besides the first survey is conducted using Google search to get the most
stand-alone model, there is ensemble classifier that combines popular database that contains these subjects. In the second
several models to gain performance improvement. The first step, the database is selected based on their reputation. The
implementation of ensemble method on epitope prediction is selected database are: IEEE explore, NCBI, PLOS, Springer
EPMeta by [26]. This server combines EPSVR, EPCES, link, and ScienceDirect.
EPITOPIA, SEPPA, PEPITO, and Discotope1.2. Hu et al. [27] The next step after search strategy is the study selection.
apply ensemble machine learning with several alternatives of Study selection conduct using inclusion and exclusion criteria
meta-learning. as follows:

III. METHODOLOGY Inclusion criteria:


• Study discussing epitope identification, epitope analysis,
Following the guidance by Kitchenham and Charters [7] epitope characteristics, and conformational epitope
this systematic literature review conduct in three stages: prediction method.
planning, conducting, and reporting. • Source from cited journal article.
• Study with strong validation and comparison with others
A. Planning popular method is included.
The need for systematic review in this research has been • The primary study which is published on 2005 or later.
discussed in the Introduction Section. The research question is • Prediction method uses data 3D structure of Ag-Ab
composed using PICOC criteria, as shown in Table 1. PICOC complex is included.
criteria are the extension from its origin (PICO) in the medical
guideline. In Computer Science such as in Software Exclusion criteria:
Engineering, Comparison means what is the intervention • Epitope identification outside the computational-based
being compared with and Context means what is the context in method is excluded.
which the intervention is delivered [7]. The research questions • Study without strong validation and comparison with
(RQ) of this systematic literature review are as follows: other popular method are excluded.
RQ1: What is the epitope characteristic that has been • Prediction method use feature that is extracted not only
identified? (motivation: identify epitope characteristic that has from the 3D structure are excluded.
been known).
IV. RESULT AND DISCUSSION
RQ2: What is the dominant feature of conformational epitope
prediction method? (motivation: Identify strong relevant The selection criteria applied to search results gives 5
feature) primary papers and 2 secondary papers on epitope analysis
RQ3: What kinds of method are used for conformational and 18 primary papers on conformational epitope prediction.
epitope? (motivation: identity strong and weakness of each There are 5 papers excluded from the paper on conformational
method) epitope prediction. The existence of primary papers in epitope
RQ4: What are parameters to evaluate prediction analysis means the papers on the epitope characteristics can be
performance? excluded. The work of Hubbard et al.[11] and Zhang et al.
(motivation: identify parameters to evaluate method [28] are categorized as secondary paper. Hubbard introduces
performance). the conformational parameter in native protein structure which
is used as reference in conformational epitope analysis, where evolutionary conservation aspect, the difference result can be
epitope is a special kind of protein-protein interaction. found at [12,9]. Kringelum et al. [11] shown that there is no
Epitope and mimotope have some similar and different significant deviation on amino acid preference between
characteristics[28]. Based on Zhang et al.[28], 5 papers on epitope and non-epitope on the same level of surface
mimotope-based conformational epitope prediction are exposure. The different result from the previous research is the
excluded from the primary paper in conformational epitope effect of difference definition of non-epitope residue.
prediction. Andersen et al.[1] define non-epitope residue as all non-
interacting residue on the antigen. Ofran et al. [30] define non-
A. Epitope characteristics epitope residue as all residue at PDB. Rubinstein et al. [3] just
RQ1: What is the epitope characteristic that have been use non-epitope residue from residue with RSA more than 5%.
identified? Kringelum et al. [11] use only non-epitope from the exposed
residue from the same exposure level. For evolutionary
There are several aspects which are analyzed in epitope conservation value, Zheng et al. [12] shows that the
characteristics: physicochemical, structure, geometric, evolutionary conservation is not significantly lower than
discontinuity level, and deep level. The analysis is conducted exposed residue non-epitope. Sun et al. [9] use AAIndex as
after filtering process. There is no standard on filtering identification parameter of epitope characteristic. AAIndex
criteria. Although the threshold used differs between papers, is proposed by Kawashima et al. [31] consist of 544 indexes. It
the general criteria are (1) Use antigen which is interacted is categorised in 6 categories: helix and turn, sheet
with the antibody (2) remove redundancy (3) data resolution conformation, residue composition, physicochemical,
(4) number of residue on the antigen. Although it is generated hydrophobic and other characteristics. Sun et al. [9] shows that
some dataset, they share many complexes. The identification as much as 21 residue characteristic at AAIndex is proven can
on Rubinstein et al. [3] and Kringelum et al. [11] shows 74% distinguish epitope from non-epitope exposed residue.
data on [3] are used in [11]. A conformational epitope is Table 3. Residual level characteristic of conformational epitope (without
composed of a number of residues that bind to the antibody, so contrary fact)
the epitope residue has two membership level: as residue Characteristic Analysis result Ref.
component and patch member. Epitope also has a unique Accessibility -Exposed on surface [32]
interaction pattern with the antibody, compared with the other -on planar area [3]
protein-protein interaction. Sun et al. [9] categorized the -epitope exist on outer layer of [12]
convex hull and deeper than exposed
parameters to describe the epitope in four perspectives: non-epitope
general characteristics, residual, structural, and interaction Residue pair Pairs of Tyr:Tyr, Cys:Pro, Asn:Tyr, [3]
patterns with an antibody. By considering the membership preference Gly:Tyr, Asp:Pro, Thr:Tyr and
type, these four categories is simplified into three categories. Arg:Tyr dominate in epitope
The general characteristics and structural characteristics of the compare to non-epitope antigen
epitope is at the patch level, so in this study, both parameters surface.
Pairs of Asn:Tyr, His:Tyr and [9]
are categorized as the characteristics of the patch level. Table His:Met dominate epitope
2 shows the identification parameters in each category. A Physicochemical AAIndex of epitope residue and non [9]
detailed explanation of each category is discussed in the next characteristics: epitope surface residue have
sub-section. AAIndex correlation value 0.8 (high
correlation)
Table 2. Identified parameter on each category Secondary Enrich by loop, depleted strand and [3]and [30]
Categorization Identified Parameter structure helix
Residual Characteristics Residue accessibility Polarity Compare to non-epitope residue, [1] [30] [3]
Amino acid Preference epitope residue contain more polar [33][9] [11]
Residue-pair preference and charge amino acid and depleted [34]
AAIndex of hydrophobic amino acid.
Secondary structure Charge Compare to non-epitope residue, [1], [30],
Evolutionary conservation epitope residue contain more charged [3], [33]and
Patch level Characteristics: Epitope size : residue number amino acid. [9]
Sequential continuity hydrophobicity Compare to non-epitope residue, [1], [30],
Geometrics shape epitope residue contain less [3], [33]
Interaction pattern with antibody Epitope-Paratope residue pair hydrophobic amino acid. and [9]
preference
Hydrophobic residue closer to
antibody while hydrophilic core [11]
flanked by charged residue
B. Residual characteristics
Most of the analysis results show their relevance with the
previous research, as shown in Table 3 and lead to specifics C. Epitope characteristics at patch level
characteristics, except for amino acid preference and Detail epitope characteristic at patch level is shown in
evolutionary conservation which shown the existence of Table 5. Generally, the analysis result at patch level shows
contrary fact as shown in Table 4. For the amino acid there is no contrary result. For example, although the filtering
preference, there is a difference between analysis result on criteria on residue number in Sun et al. [9] are greater than in
Kringelum et al. [11] with the previous result in [9,29]. In the Kringelum et al. [11], the analysis result shows that range of
residue number at [11] is covered by [9]. This result shows the exposed level of residue. There is some measure can be
that antigen size does not influence the residue number of used to represent this characteristic: Accessible Surface Area
epitope. From geometric aspect, result analysis shows the (ASA), Relative Surface Area (RSA), Half Sphere Exposure
more specific characteristic of epitope can be identified. (HSE), and Contact Number (CN). The features derived from
Table 4. Residual level characteristic of conformational epitope
solvent accessibility which is extracted first in the epitope
(with contrary fact) prediction method are ASA and RSA. These characteristics
Characteristic Analysis result Reference can separate exposed residue from buried residue. Other
Amino acid Contain more Trp, Tyr, Arg and [9] features derived from other characteristics, such as
preference * His and less Cys, Ala and Val physicochemical, geometrics, and evolutionary conservation
Contain more Gly, Trp, Tyr, His, are used to calculate antigenicity score of each exposed
Gln, Met [34]
Contain more charged residue:
residue. The list of epitope characteristics which is used in
Asp, Glu, Lys, Arg, His; non- [12] prediction method is shown in Table 5.
charge residue: Asn, Gln; Table 5. Epitope characteristics in prediction method
aromatic residue: Tyr
Trp and Tyr dominate the Considered epitope characteristics Ref
antigen surface. Accessibility (CN) 1,16
Can’t distinguish between [29] Statistic deviation between epitope 1,14,38
epitope residue and non-epitope and non-epitope
exposed residue but some [11]* Accessibility (HSE) 13,14,
residue is rarely appear. Accessibility (ASA ) 37,17,18,22,21
evolutionary Less conserve than exposed [9] Geometric properties 37
conservation* residue non epitope Patch level characteristic 16,17,18,
Physicochemical 18,19,38,21,22
Conservation score not Accessibility (RSA ) 1,16,20,22
significantly difference than [12]
structural- geometrical properties 19,20,38,22,
exposed residue non epitope
Amino acid composition 20
Evolutionary conservation 20,21,38,
Conformational epitope composed of spatially neighboring Flexibility 22
Amino acid residue pair 21
residues which are separated in the primary structure or
sequence. From the analysis result, it is shown that the
conformation is composed of linear segments with variation The relation between epitope characteristics is still
length. Sun et al. [9] states that these linear segments may unknown, some researchers use statistic deviation and
have a definite role in the antibody binding. This statement represent this characteristic as propensity score. The first
needs further analysis. In windows-based sequential analysis attempt is conducted by Andersen et al.[1] which propose the
[10] both epitope and functional epitope are scattered on the log-odd ratio calculated from the epitope dataset. Andersen’s
sequence. This finding makes windows-based prediction log-odd ratio is derived use Gibbs sampling method proposed
method on the primary structure of protein is reasonable to in [39]. Kringelum et al. [14] combine the Andersen’s idea
detect linear segments of B-cell epitope. with Sweredoski and Baldi’s idea [13] and propose log-odd
ratio in spatial neighborhood context. Kringelum work [14]
D. Epitope characteristics on Ag-Ab Interaction shows that successful attempt to bring epitope characteristics
Analysis on Ag-Ab interaction is still rare nevertheless in spatial context can improve the performance of the method.
Ag-Ab interaction is specific than other protein interaction, as RQ3: What kinds of method are used for conformational
reinforce by Jackson [35] where interaction between antigen epitope?
and antibody occur at side chain not in the main chain. One of
the reasons for this scarcity is researcher’s focus on epitope Identification on existing conformational epitope
exclusively in the beginning so that the interaction with prediction method shows that in the beginning, the prediction
antibody is ignored. Several characteristics have been method ignores imbalance class problem and focuses on the
identified are its relative orientation to antibody, Contact with developing of discriminant function on the basis of feature
CDR, AA Preference on CDR, and B Factor. According to combination. The proposed method that use this approach is
[11], epitope orientation is -30o to 60o to light and heavy score which is derived from statistic deviation [1] is
chain antibody. Rubinstein et al. [3] show that 90% area comparable with the methods which is combine several
epitope contact with CDR. CDR contain more Ser and Tyr known epitope characteristics, as shown in [16]. Bringing the
with Tyr create more Hydrogen binding, and beta factor on propensity score in spatial context as in [14 improves the
CDR area is lower than on non-CDR area [36]. performance significantly. Unit patch of residue triangle [17]
is better in describing the local spatial context. Embedding the
E. B-cell conformational epitope prediction methods epitope characteristic at local spatial context [18] improves the
performance. Applying non-linear classifier such as SVM,
RQ2: What is the dominant feature on conformational epitope ANN, and Gaussian Mixture at [13] by ignoring imbalance
prediction method? class does not gain optimal performance. It is caused by the
The dominant feature on conformational epitope is feature machine learning weakness such as SVM on imbalance class
that represents solvent accessibility. This feature represents problem that is not handled properly.
Further development of epitope prediction methods, where antibody interaction phenomena. The knowledge about
the feature evolves to large dimension, the machine learning antigen-antibody interaction opens room for the development
approach is more successful than the previous attempt. of graph-based prediction method such as in [3,11, 36].
Several approaches consider the imbalance class problem,
such as in [20, 21, 22]. Three approaches are used to resolve RQ4: What parameters are used to evaluate prediction
imbalance class problem, i.e. data level, algorithm level and performance?
hybrid. On the data level, the method is applied to achieve There is no agreed upon standard to measure the performance
balances in both positive class and negative class of data of prediction method. Researchers use variety combination of
training. On Zhang et al.[20] imbalance class problem is the area under the receiver operating characteristic (ROC)
solved on data level using bootstrapping to down sample the curve (AUC), accuracy, recall, specificity, sensitivity,
negative class randomly. Down sampling approach in precision, F-Score and MCC. Epitope prediction method is a
imbalance class potentially eliminates the key feature. On classification problem. In the classification problem, the
algorithm level, there are two approaches. The first approach standard method to measure general performance is the area
uses algorithm modification to reduce bias to the majority under the receiver operating characteristic (ROC) curve
group. Zhang et al. [21] propose cost-sensitive boosting to (AUC)[4]. On imbalance class, AUC underestimates the actual
develop strong classifier from the weak classifier. The second predictive power because it considers prediction that is not
approach implements one class learning that focuses on the part of any validated epitope as false prediction [3]. Accuracy
target group. Ren et al. [22] use algorithm level approach that (recognition rate) describe recognition rate both positive and
use one class classifier method, i.e. positive unlabeled negative sample and effective on balance sample [22]. On
learning. imbalance class, the majority class will achieve better
On further developments, along with the accumulated accuracy. Recall, specificity, and precision reflect the
knowledge of the epitope, the number of feature vector prediction tendency of the classifier. F-Score is a combination
developed rapidly. Other than imbalance class problem, of precision and recall and effective to measure both on
feature selection is one of the open problems to build effective balance and imbalance sample. MCC is effective on imbalance
classifier on epitope prediction. First attempt to use feature sample with the range from +1 to -1 by indication: +1: perfect
selection is conducted by Rubinstein [19]. He uses top down prediction, 0: random prediction, -1: totally reverse prediction.
heuristic searches to select 44 immunogenic characteristics. As resume of the existing approach, there is still a limited
Zhang et al. [20] use Fisher-Markov Selector and incremental number of methods proposed to solve the conformational
Feature Selection to find optimal feature. Considering feature epitope prediction. There are some advances have been
representation and imbalance class problem to solve epitope achieved on this field of study, but there are still chances to
prediction gives better model, as shown at [22] where their improve the performance. Considering imbalance class
approach outperformed SEPPA 2.0, Discotope 2.0 and Ellipro. problem, doing feature selection, improving the features and
The main limitation of this approach is that the feature must integrating them to make strong classifier are main keys to
have been known at the beginning of the process. This make effective prediction.
condition is not satisfied in real condition, where it is possible
to have new feature after the selection process is ended. V. CONCLUSSION AND FUTURE WORK
Besides the stand-alone model, there is ensemble classifier This study identifies the relationship between epitope
that combines several models to gain performance characteristic and the performance of prediction method.
improvement. The performance of ensemble model is the best There are two research areas that have strong relation with this
between the combined model, but still can not achieve the study, research on epitope analysis and research on
desirable performance. The first implementation of ensemble conformational epitope prediction. There are some epitope
method on epitope prediction is EPMeta [26]. This server characteristics have been identified in epitope analysis.
combines EPSVR, EPCES, EPITOPIA, SEPPA, PEPITO, and However, there is no proposed method that considers all of the
Discotope1.2. Hu et al.[27] apply ensemble machine learning characteristics. This is caused by the difficulty to extract
with several alternatives of meta-learning to identify the feature on 3D structure base. To achieve good performance on
relation between the accuracy gain and epitope characteristics epitope prediction there are three main problems must be
which is considered in prediction method, in this article the solved, i.e. the imbalance class problem, feature extraction,
performance of the method is assessed based on characteristics and the feature selection and integration. The specificity of
completeness. The machine learning approach has been epitope-paratope interaction also makes a chance to make a
improved significantly. However, it still has limitations on new approach in epitope prediction or improve the graph
proposed method and huge room for performance based approach. The number of data sets relatively fixed, but
improvement exists [24]. the feature continues to evolve in line with the new finding on
Other approach that has been applied to epitope prediction feature representation of epitope. This condition has new
method is a graph-based method, as proposed in [24,25]. Zhao problem in epitope prediction: if there are new features, are
et al.[24] use Markov Clustering graph algorithm to these features can be incorporated into the classifier.
distinguish epitope from non-epitope. Zhao et al. [25] use Considering imbalance class problem and feature selection on
graph mining to find the coupling sub graph and propose developing algorithm make a chance to improve the accuracy
graph transformation to transfer coupling graph to generic of prediction gradually.
graph. The coupling graph technique is in line with antigen-
[20] W. Zhang, Y. Xiong, M. Zhao, H. Zou, X. Ye, and J. Liu, “Prediction of
References conformational B-cell epitopes from 3D structures by random forests
with a distance-based feature,” BMC Bioinformatics, vol. 12, no. 1, p.
341, 2011.
[1] P. H. Andersen, M. Nielsen, and O. L. E. Lund, “Prediction of residues
in discontinuous B-cell epitopes using protein 3D structures,” pp. 2558– [21] J. Zhang, X. Zhao, P. Sun, B. Gao, and Z. Ma, “Conformational B-Cell
2567, 2006. Epitopes Prediction from Sequences Using Cost-Sensitive Ensemble
Classifiers and Spatial Clustering,” vol. 2014, 2014.
[2] H. M. Berman, J. Westbrook, Z. Feng, G. Gilliland, T. N. Bhat, H.
Weissig, I. N. Shindyalov, and P. E. Bourne, “The protein data bank.,” [22] J. Ren, Q. Liu, J. Ellis, and J. Li, “Positive-unlabeled learning for the
Nucleic Acids Res., vol. 28, no. 1, pp. 235–242, 2000. prediction of conformational B-cell epitopes,” BMC Bioinformatics, vol.
16, no. Suppl 18, p. S12, 2015.
[3] N. D. Rubinstein, I. Mayrose, D. Halperin, D. Yekutieli, J. M. Gershoni,
and T. Pupko, “Computational characterization of B-cell epitopes,” Mol. [23] U. Kulkarni-kale, S. Bhosle, and A. S. Kolaskar, “CEP : a
Immunol., vol. 45, no. 12, pp. 3477–3489, 2008. conformational epitope prediction server,” vol. 33, pp. 168–171, 2005.
[4] J. V Ponomarenko and M. H. V van Regenmortel, “B-Cell Epitope [24] L. Zhao, L. Wong, L. Lu, S. C. H. Hoi, and J. Li, “B-cell epitope
Prediction BT - Structural Bioinformatics,” Struct. Bioinforma., p. prediction through a graph model.,” BMC Bioinformatics, vol. 13 Suppl
1064, 2009. 1, no. Suppl 17, p. S20, 2012.
[5] P. Sun, H. Ju, Z. Liu, Q. Ning, J. Zhang, X. Zhao, Y. Huang, Z. Ma, and [25] L. Zhao, S. C. H. Hoi, Z. Li, L. Wong, H. Nguyen, and J. Li, “Coupling
Y. Li, “Bioinformatics resources and tools for conformational B-cell graphs, efficient algorithmsand B-cell epitope prediction,” IEEE/ACM
epitope prediction,” Comput. Math. Methods Med., vol. 2013, 2013. Trans. Comput. Biol. Bioinforma., vol. 11, no. 1, pp. 7–16, 2014.
[6] I. Sela-Culang, Y. Ofran, and B. Peters, “Antibody specific epitope [26] S. Liang, D. Zheng, D. M. Standley, B. Yao, M. Zacharias, and C.
prediction - Emergence of a new paradigm,” Curr. Opin. Virol., vol. 11, Zhang, “EPSVR and EPMeta : prediction of antigenic epitopes using
pp. 98–102, 2015. support vector regression and multiple server results,” 2010.
[7] B. Kitchenham and S. Charters, “Guidelines for performing Systematic [27] Y.-J. Hu, S.-C. Lin, Y.-L. Lin, K.-H. Lin, and S.-N. You, “A meta-
Literature reviews in Software Engineering Version 2.3,” Engineering, learning approach for B-cell conformational epitope prediction.,” BMC
vol. 45, no. 4ve, p. 1051, 2007. Bioinformatics, vol. 15, no. 1, p. 378, 2014.
[8] S. J. Hubbard, R. J. Beynon, and J. M. Thornton, “Assessment of [28] C. Zhang, Y. Li, W. Tang, Z. Zhou, P. Sun, and Z. Ma, “The
conformational parameters as predictors of limited proteolytic sites in Relationship between B-cell Epitope and Mimotope Sequences.,”
native protein structures.,” Protein Eng., vol. 11, no. 5, pp. 349–359, Protein Pept. Lett., vol. 23, no. 2, pp. 132–41, 2016.
1998. [29] T. Ramaraj, T. Angel, E. A. Dratz, A. J. Jesaitis, and B. Mumey,
[9] J. Sun, T. Xu, S. Wang, G. Li, D. Wu, and Z. Cao, “Does difference “Antigen-antibody interface properties: Composition, residue
exist between epitope and non-epitope residues? Analysis of the interactions, and features of 53 non-redundant structures,” Biochim.
physicochemical and structural properties on conformational epitopes Biophys. Acta - Proteins Proteomics, vol. 1824, no. 3, pp. 520–532,
from B-cell protein antigens,” Immunome Res., vol. 7, no. 3, pp. 1–11, 2012.
2011. [30] Y. Ofran, A. Schlessinger, and B. Rost, “Automated identification of
[10] G. N. Sivalingam and A. J. Shepherd, “An analysis of B-cell epitope complementarity determining regions (CDRs) reveals peculiar
discontinuity,” Mol. Immunol., vol. 51, no. 3–4, pp. 304–309, 2012. characteristics of CDRs and B cell epitopes.,” J. Immunol., vol. 181, no.
9, pp. 6230–6235, 2008.
[11] J. V. Kringelum, M. Nielsen, S. B. Padkjær, and O. Lund, “Structural
analysis of B-cell epitopes in antibody: Protein complexes,” Mol. [31] S. Kawashima, P. Pokarowski, M. Pokarowska, A. Kolinski, T.
Immunol., vol. 53, no. 1–2, pp. 24–34, 2013. Katayama, and M. Kanehisa, “AAindex : amino acid index database ,
progress report 2008,” vol. 36, no. November 2007, pp. 202–205, 2008.
[12] W. Zheng, J. Ruan, G. Hu, K. Wang, M. Hanlon, and J. Gao, “Analysis
of conformational B-cell epitopes in the antibody-antigen complex using [32] J. M. Thornton, M. S. Edwards, W. R. Taylor, and D. J. Barlow,
the depth function and the convex hull,” PLoS One, vol. 10, no. 8, pp. 1– “Location of ‘continuous’ antigenic determinants in the protruding
16, 2015. regions of proteins.,” EMBO J., vol. 5, no. 2, pp. 409–413, 1986.
[13] M. J. Sweredoski and P. Baldi, “BIOINFORMATICS APPLICATIONS [33] L. Zhao and J. Li, “Mining for the antibody-antigen interacting
NOTE Structural bioinformatics PEPITO : improved discontinuous B- associations that predict the B cell epitopes,” Struct. Biol., vol. 10, no.
cell epitope prediction using multiple distance thresholds and half sphere Suppl 1, pp. 1–13, 2010.
exposure,” vol. 24, no. 12, pp. 1459–1460, 2008. [34] Y. T. Lo, T. W. Pai, H. H. Hsu, and H. K. Tsai, “Epitope and Paratope
[14] J. V. Kringelum, C. Lundegaard, O. Lund, and M. Nielsen, “Reliable B Region Analysis,” 2014 Eighth Int. Conf. Complex, Intell. Softw.
Cell Epitope Predictions : Impacts of Method Development and Intensive Syst., pp. 510–514, 2014.
Improved Benchmarking,” vol. 8, no. 12, 2012. [35] R. M. Jackson, “Comparison of protein-protein interactions in serine
[15] A. Schlessinger, Y. Ofran, G. Yachdav, and B. Rost, “Epitome: database protease-inhibitor and antibody-antigen complexes: implications for the
of structure-inferred antigenic epitopes.,” Nucleic Acids Res., vol. 34, protein docking problem.,” Protein Sci., vol. 8, pp. 603–613, 1999.
no. Database issue, pp. D777-80, 2006. [36] T. Osajima, M. Suzuki, S. Neya, and T. Hoshino, “Computational and
[16] S. Liang, D. Zheng, C. Zhang, and M. Zacharias, “Prediction of statistical study on the molecular interaction between antigen and
antigenic epitopes on protein surfaces by consensus scoring,” BMC antibody,” J. Mol. Graph. Model., vol. 53, pp. 128–139, 2014.
Bioinformatics, vol. 10, no. 1, p. 302, 2009. [37] J. Ponomarenko, H. Bui, W. Li, N. Fusseder, P. E. Bourne, A. Sette, and
[17] J. Sun, D. Wu, T. Xu, X. Wang, X. Xu, L. Tao, Y. X. Li, and Z. W. Cao, B. Peters, “ElliPro : a new structure-based tool for the prediction of
“SEPPA: A computational server for spatial epitope prediction of antibody epitopes,” vol. 8, pp. 1–8, 2008.
protein antigens,” Nucleic Acids Res., vol. 37, no. SUPPL. 2, pp. 612– [38] J. Ren, Q. Liu, J. Ellis, and J. Li, “Tertiary structure-based prediction of
616, 2009. conformational B-cell epitopes through B factors,” Bioinformatics, vol.
[18] T. Qi, T. Qiu, Q. Zhang, K. Tang, Y. Fan, J. Qiu, D. Wu, W. Zhang, Y. 30, no. 12, pp. 264–273, 2014.
Chen, J. Gao, R. Zhu, and Z. Cao, “SEPPA 2.0 - More refined server to [39] M. Nielsen, C. Lundegaard, P. Worning, C. Sylvester Hvid, K.
predict spatial epitope considering species of immune host and Lamberth, S. Buus, S. Brunak, and O. Lund, “Improved prediction of
subcellular localization of protein antigen,” Nucleic Acids Res., vol. 42, MHC class I and class II epitopes using a novel Gibbs sampling
no. W1, pp. 59–63, 2014. approach,” Bioinformatics, vol. 20, no. 9, pp. 1388–1397, 2004.
[19] N. D. Rubinstein, I. Mayrose, and T. Pupko, “A machine-learning B. J. Mcconkey, V. Sobolev, and M. Edelman, “atom – atom contacts
approach for predicting B-cell epitopes,” Mol. Immunol., vol. 46, no. 5, using a constrained Voronoi procedure,” vol. 18, no. 10, pp. 1365–1373,
pp. 840–847, 2009. 2002

View publication stats

You might also like