Chapter 2.2. Improving in vitro diagnostic design with in silico support: Structural bioinformatics to the fore


Yee Siew Choong, Theam Soon Lim and Eugene Boon Beng Ong

Art work
Uli Reinhardt
What happens when corn and wheat prices rise is that we see real increases in malnutrition and under-nutrition. 
And when children are malnourished, their brain development actually slows down and is affected. 
So, this is not just a short-term impact. Jim Yong Kim
Who will save this little child,
smaller than an oat grain?
From where will come the hammer
executioner of this chain?
Miguel Hernández
The child of the plough


The great advances in data management and mining technologies over the past decade have brought about huge changes in many fields of research. Improvements in software and hardware mean easier and faster management, extraction, and interpretation of biological data. Bioinformatics relies heavily on the systematic integration and connection of huge collections of data by information technology to translate higher-order information. In this way, value is added to large volumes of biological information such as microarray data, gene/protein sequences, and the three-dimensional structures of biomacromolecules and ligands obtained from experimental data when the sequence–structure–function relationship is forged. In light of this, it is necessary to visualize and mine large quantities of biological data to define new patterns and search for novel candidates in biomedical research. Advances in computational technology and resources allow such massive collections of data to influence structural bioinformatics. Structural bioinformatics facilitates the simulation and modeling of proteins to gain a deeper understanding of the properties and behaviors of a given system. In diagnostic applications, computational structural biology provides a new perspective in the design formulation of ligands for the detection of a specific target. This is evident in the many in silico approaches used to define antibody–antigen interactions, where the intermolecular network can be observed in detail. This chapter provides a review of relevant published works and provides an overview of the impact of structural bioinformatics on the field of diagnostics research, focusing mainly on antibody–antigen interactions (Figure 1).

Figure1. General flowchart depicting the application of structural bioinformatics in diagnostics.

Structural Bioinformatics

Structural bioinformatics allows the conceptualization of biology and focuses on the analysis and prediction of the three-dimensional structures of biological macromolecules and their structure–function relationships. It has been catalyzed by the current availability of huge amounts of structural and genomic data. X-ray crystallography, nuclear magnetic resonance (NMR) spectroscopy, and cryo-electron microscopy contribute to structural information, and genomic data from DNA/RNA sequencing, mass spectrometry, and microarray expression analysis provide more information for predictions. Phenotypic and genotypic information can be correlated and linked in biological phenomena, and computational technology and mathematical calculations can be used to obtain a deeper understanding of how cellular pathways and processes work within cells to regulate certain biological functions associated with life.

The increased number of protein structures and the development of better modeling methodologies have steadily improved the reliability of predicted protein structures from sequence information. Improvement in the quality of three-dimensional protein models has thus led to their routine use in proteomics research. Protein structure prediction, a key bioinformatics technique, emerged as a method for rapidly deriving the three-dimensional structure of a protein from its sequence. Other essential methods that are dependent on protein structure prediction are molecular docking simulation and molecular dynamics simulation, which can be used to study interactions between molecules and for epitope/paratope prediction. Thus, properties such as serum half-life, binding affinity, solubility, and protein or antibody activity can be optimized.

Protein Structure Prediction

Proteins are biomacromolecules that are responsible for almost every activity within a cell. Biological functions that are of immediate medical relevance include metabolic reactions, transport/carrying, cell signaling, defense mechanisms, and responses to stimuli. Such reactions are mediated either by a single protein or by a combination of proteins that interact through a cascade. This is because most of the biological functions of a protein are dictated by its conformation, either independently or in association with another protein in complex. For some cell-mediated responses, the unique complexation of several proteins is the key regulatory factor for certain cell-based mechanisms.

To fully understand the function of a protein and its role in network interactions, knowledge of its three-dimensional structure is essential. The unique characteristics of the various amino acids and differences in the length of a protein’s primary sequence result in differences in the final three-dimensional structure. Even taking into account the obvious importance of protein three-dimensional structure data, the number of actual protein structures being solved using laboratory methods is still low. The obvious problems associated with protein crystallization make solving structures a huge challenge, even for the best-trained professionals. However, owing to the discrepancy between the number of solved protein structures and the number of non-redundant protein sequences, protein structure prediction offers a suitable alternative when the various experimental approaches have failed to solve the three-dimensional structure of a given protein. Protein structure prediction as an alternative to conventional laboratory methods for the determination of protein structural features is especially important when the protein is also an antigen or a biomarker of a certain disease. The ability of bioinformatics tools to successfully predict the structure of an antigen is key to the design of ligands such as antibodies, which can tailored for specific antigens. Structure prediction approaches range from methods that require very little computational power, such as template-based comparative modeling, to template-free ab initio modeling, which requires more intensive computing resources.

Comparative modeling involves several general steps: template identification, sequence alignment, framework construction, loop and side-chain modeling, and evaluation of the final model. It is based on two main observations: (1) the tertiary structure of a protein can be obtained from its primary structure because the protein’s structure is determined by its primary amino acid sequence [1]; and (2) proteins with similar sequences will adopt identical structures, and distantly related sequences will still fold into similar structures [2-4]. Hence, when the sequence identity between the target and template is within the tolerance limit of proximity (i.e., 30% sequence identity with the template) [3], comparative modeling remains the most accurate structure prediction method. In general, with more than 50% sequence identity with the template, comparative modeling can produce an accurate overall structure with a 1.0 Å root mean square (RMS) error in backbone atoms, whereas 90% of backbone atoms will have 1.5 Å RMS error when the sequence identity is 30–50% [5, 6]. For models built below the sequence identity threshold, which is less than 30%, the structures will tend to have greater alignment errors. Errors in comparative modeling are usually associated with side-chain positioning, core distortion, loop modeling, and alignment discrepancies, which can result in failure of ligand design [7], even when the overall structure has a high degree of accuracy. It is important to remember that even with such high accuracy in the overall structure, sufficient accuracy of side-chain positioning is also vital, and can affect drug design, as reported by DeWeese-Scott and Moult (2004), because side chain conformational sampling is important when performing docking predictions [7]. However, a comparative models of G protein-coupled receptor [8] and solute carrier transporter [9], which were built based on sequence identities of less than 25% with the template, could still be useful providing model optimization considerations are taken into account. Therefore, even with sequence identities of 25%, comparative modeling can be carried out successfully and can yield useful information [10].

Ab initio or de novo structure prediction focuses on the use of only the amino sequence to predict the conformation of a protein with the assumption that most proteins at or near to the minimum global free energy are in the native conformation [11, 12]. Ab initio methods predict a protein’s structure by conformation space sampling guided by scoring function and sequence-dependent biases to generate a large set of “decoy structures” for the selection of “native” conformations. The scoring functions can be knowledge-based, where the properties of native conformations are obtained from statistical models, and/or physics-based, where the molecular interactions are described using mathematical models.

Because ab initio modeling requires extensive computational resources to identify the lowest energy of a protein in its native state, insufficient conformational sampling would therefore be a major bottleneck in this template-free protein modeling approach. In terms of model accuracy, ab initio methods are only successful for short peptides (i.e., root mean square deviation (RMSD) of approximately 1.0 Å for a 20-residue Trp-cage peptide) [13-15], and small proteins of less than 100 residues (RMSD of approximately 2.0–5.0 Å) [16-22].

In between the template-based and template-free modeling techniques mentioned above are hybrid/informed methods that utilize statistical or data mining approaches to determine the structural classes of proteins from their amino acid sequences using numerous learning algorithms. These machine-learning techniques recognize data and summarize them into patterns. Artificial neural networks (ANNs), hidden Markov models (HMMs), support vector machines (SVMs), and blended multi-method algorithms are the main approaches used for this kind of data training. In principle, ANNs build structural summaries from data and patterns when a sufficiently large number of processing elements can approximate any continuous functional mapping with some level of accuracy [23, 24].

HMMs predict proteins structures using Markov states for the presentation of homologous sequence positions, secondary structure types, or membrane locality where they contain the position-specific probabilities for inserts and deletions along the multiple sequence alignment [25, 26]. In contrast, SVMs are able to handle non-linear classification, and are thus more efficient at determining the non-linear sequence–structure relationship [27-30]. Blended multi-method algorithms use a combination of other methods to balance the shortcoming of the integral methods. Bouziane et al. (2011) [31] combined ANNs, multiclass-SVMs, and K-NNs (k-nearest neighbors), and found that this ensemble combination method yielded significant improvement in the accuracy of protein secondary structure prediction compared with individual classifiers. Zou et al. (2010) [32] also showed that a combined HMM and genetic algorithm (GA) approach is the best choice for predicting transmembrane topology and the signal peptide of β-barrel proteins because the model can be trained more quickly and deploys parallel computing for faster calculation.

 Antigen Structure Prediction

In the context of antibody–antigen interactions, an antibody functions as the binding entity, whereas an antigen, usually a protein, is the captured target. The mechanism that brings about in vivo antibody production is multifaceted and involves several immunological responses. Antigens, including autoantigens, are recognized as foreign bodies by the host and elicit the production of antibodies by B-cells against these foreign targets. From a structural perspective, the paratope on an antibody generates a core within its structure that complements the antigen surface region commonly known as the epitope. The specificity of an antigen primarily results from its unique three-dimensional conformation, which is recognized by a specific antibody. Elucidation of the structure of an antigen is therefore vital for disease diagnosis and prognosis, drug and vaccine development, and an understanding of the cellular pathways of the disease. The three-dimensional structure of antigens can therefore be used to test hypotheses about their function using mutants, and for active site identification, epitope identification, lead optimization of inhibitors, and investigations into antigen–antibody complex interactions for antibody design to improve antibody-binding affinity, etc.

The application of such techniques has been used in the diagnosis of diabetes mellitus type I. Glutamic acid decarboxylase (GAD 65) is an autoantigen that is relevant to diabetes mellitus type I because it plays an important role in the regulation of glucagon secretion [33-35]. Wiltgen et al. [36] obtained the structure and identified the active site of GAD 65 via homology modeling and sequence alignment with the template structures. Their predictive models revealed the antigenic determinants of GAD 65, which is potentially useful information for the design of antibodies for diagnosis, prognosis, vaccination, and therapy against diabetes mellitus type I.

The diagnosis of hepatitis B virus (HBV) infection is commonly based on an antibody detection assay against hepatitis B surface antigen (HBsAg). However, failure of HBsAg detection in HBV-infected individuals has been reported [37-39]. It is thought that the failure was caused by a change in the three-dimensional conformation of HBsAg at the “a” determinant variants, which are important for antibody recognition [40-42]. Because the tertiary structure of HBsAg has not yet been fully determined, Ie et al. [42] performed ab initio modeling to predict the structure of HBsAg. From the predicted structures, they showed that T143M has a larger shift in the second loop of HBsAg that contributes to the most significant antigen index changes. Their study highlighted that the antigen’s structural alterations need to be taken into account during HBV screening. This finding suggests that the prediction of an antigen’s structure is possible and can be used to improve particular antibody-based diagnostics. The research also shows that the structural alteration of epitopes needs to be taken into consideration to improve the sensitivity of diagnostics and the development of vaccines and immunotherapies. Therefore, protein structure prediction is a useful tool that can provide information to overcome detection failure.

In a study on the hepatitis G virus (HGV) or GB virus C (GBV-C), Ranjbar et al. [43] evaluated the characteristics, mutations, structures, and antigenicity of the E2 envelop protein. They combined several bioinformatics approaches to study the computational molecular features and immunoinformatics characteristics of the E2 envelop glycoprotein. The data extracted from the three-dimensional structure of the E2 envelop showed residues 231–296 to be optimal for immunization and diagnostic development, whereas residues 215–223 could be used as the T-cell epitope.

The cell membrane isolates a cell from the external world and transmembrane proteins allow the cell to communicate via extracellular stimuli acting on the cell. The analysis of solved membrane protein structures using computational methods will have a great impact on membrane protein research. The structure of such insoluble proteins can be used in antibody design for more direct disease diagnosis, i.e., antigen-based diagnosis to detect pathogens. Moreover, drug discovery can also be applied to screen for possible inhibitors of efflux pump or ion channel membrane proteins. Predictive models of outer membrane proteins (OMPs) from Salmonella enterica serovar Typhi [44] and Shigella flexneri [45] provide additional information about the proteins as well as their suitability for diagnostics or therapy. Comparative models obtained from OMPs from S. enterica ser. Typhi [44] and S. flexneri [45] were used to identify potential epitopes to design specific binders useful for antigen-based diagnostics.

Owing to the lack of a precise rapid diagnostic test, Rahbar et al. conducted an in silico analysis of the antigenic conserved regions of biofilm-associated proteins (Baps) specific to Acinetobacter baumannii [46]. An antigenic conserved region of Bap was selected based on alignments and propensity scales, and a three-dimensional structure of the region was built for B-cell epitope mapping. The protein was found to be a potential antigen with several antigenic determinants that were capable of eliciting antibody responses. Therefore, the specific Bap protein was suggested as a suitable agent for antibody–antigen-based diagnostics, as well as for the exploration of possible antibodies to protect individuals at risk from nosocomial A. baumannii infection.

 Antibody Structure Prediction

An antibody molecule is a Y-shaped polypeptide chain consisting of two identical heavy chains and two identical light chains connected by disulfide bonds. Antibodies are classified according to their heavy chains, i.e., either α, γ, ε, μ, or δ. Each heavy chain consists of constant and variable regions. The light chain of an antibody is designated either λ or κ. The length and amino acid composition of an antibody are the key features of its structure. The prong of the Fab (fragment, antigen-binding) region in an antibody is the antigen-binding site. This site is often called the paratope and it binds to a specific epitope on an antigen with high selectivity because the variable domains dictate its particular shape. The Fc (fragment, crystallizable) region of an antibody is a constant region and is important for modulating immune cell activity. Next-generation sequencing technology can now provide information about the length and composition of the variable regions at high throughput. Variations in the length and composition of the chains are the basis of the large repertoire of antibodies generated by the immune system.

These sequences become very useful when designed antibodies are required. The length of the complementarity-determining regions (CDRs) is related to the physical structure of the target antigen to be captured. The variable region of the heavy chain is encoded by variable (V), diversity (D), and joining (J) gene segments, whereas the variable region of the light chain consists of only V and J segments. Therefore, V(D)J recombination is capable of yielding billions of possible combinations to generate antigen-specific antibodies. The production of a single monoclonal antibody against an antigen allows specific detection. Therefore, in a laboratory antibody-based diagnosis of a disease, the specificity of the antibody or immunoglobulin (Ig) for the antigen is vital for the accuracy of the test, and for avoiding false-positives and cross-reactions with other antigens.

In general, the canonical structure of an antibody largely depends on the framework sequences. The overall structure of antibodies might be the same but changes in the framework can result in key angle changes of the loops that are associated with antigen binding. CDR grafting studies have demonstrated that structurally divergent frameworks carry an increased risk of losing affinity, and may lead to decreased stability of antibody [47]. The findings of Honegger et al. on the context-dependence of the optimal VH framework also demonstrate why the immune system is built on a set of frameworks to support the different CDRs [48].

In an investigation of the importance of specific residues in single-chain variable fragment (scFv) design, docking simulation was performed on nine natural mutants of HIV-1 p17 with scFv anti-p17 [49]. Calculated potential mean force scores correlated well with the results of the peptide-based enzyme-linked immunosorbent assay (ELISA). The binding free energy and the results of the pairwise decomposition of scFv anti-p17 with the HIV-1 p17 complex showed that five residues in the CDRs (Met100, Lys101, Asn169, His228, and Leu229) of scFv anti-p17 were major contributors to the binding interaction, where the absolute decomposition energy is more than 2 kcal/mol. The results demonstrated that the pairwise decomposition energy calculation can identify the important residues involved in antigen binding, and mutation of these residues can potentially improve the binding affinity of an scFv.

In a subsequent work, Tue-ngeun et al. [50] used computational alanine scanning to determine if side-chain residues in CDRs play an important role in bioactivity. They studied scFv anti-p17 and scFv anti-p17 mutants with nine HIV-1 p17 mutant complexes. On the basis of this computational alanine scanning calculation and the earlier work on pairwise decomposition energy [49], Met100 at scFv anti-p17 was selected for mutation to investigate the effects of point mutation on the binding affinity of scFv for HIV-1 p17 protein. The theoretical results showed that a single point mutation at M100G/R can improve the binding affinity of scFv for anti-p17. The improvement is attributed to the electrostatic interactions compared with the wild type. The calculation was validated and agreed well with the results obtained from the peptide ELISA data (highest percentage inhibition values of 79.46% for p17.8) obtained for a newly designed scFv anti-p17 with improved HIV-1 binding.

Antibody-antigen interactions

The development of antibody engineering and the in vitro evolution of antibodies have opened up new avenues in de novo antibody design to satisfy therapeutic and diagnostic requirements. However, high-resolution protein–protein structures are one of the key factors in understanding the mechanisms of protein–protein interactions and the effects of mutations on binding affinity. More specifically, in structure-based antibody engineering, knowledge of antibody–antigen interaction enables the identification of amino acids suitable for modification or mutation, and allows prediction of the substitutions to be made. The significant impact of VL-VH relative orientation on the antigen-binding conformation of an antibody (VL is the light chain variable domain and VH is the heavy chain variable domain) [51] further demonstrates that intramolecular changes occur upon antigen binding with an antibody when both VL-VH and antibody–antigen relative orientation are optimized simultaneously. This emphasizes the importance of the study of antibody–antigen complexes at high resolution; there is a huge discrepancy between the number of solved complex structures and the number of available sequences of protein complexes owing to the time and cost required for solving these structures experimentally. Therefore, the prediction of protein–protein complexes by computational calculation, i.e., molecular docking simulation, could potentially provide a rapid, efficient, and unbiased alternative option that provides the detailed information required for antibody design.

A comparison of 49 free and bound antibodies showed that they are subject to antigen binding-related allosteric effects [52]. Analysis revealed that only one third of antibodies undergo significant binding-related conformational changes at the binding site of CDR-H3. Apart from the antigen-binding site, the most consistent and substantial conformational changes occur at a loop in the heavy chain constant domain, followed by the elbow angle between the variable and constant domains and the relative orientation of heavy and light chains. This loop is associated with the interaction between heavy and light chains, and involves complementary binding with the Fc receptor. These findings show that antibody modeling and engineering could be improved if the structural changes occurring in the antibody during antigen binding were identified.

Computational studies of antibody–antigen complexes have revealed that antigen binding causes preferential conformational changes, and intramolecular interaction is a phenomenon that is shared by all antibodies [53]. These changes can affect the binding site as well as the Fc effector region. The process of antibody–antigen recognition from a dynamic and energy perspective could potentially aid the design of antibodies that are capable of specific antigen binding in diagnostic and therapeutic applications.

Sircar and Gray have illustrated a new docking algorithm for the optimization of paratopes for antibody–antigen binding to compensate for errors in antibody homology models [54]. The study improved the accuracy of the intermolecular and interfacial flexibility of an antibody compared with standard rigid body docking. The combination of ensemble docking for conformer selection and induced fit to increase the sampling of diverse antibody conformations can help improve antibody design.

Molecular dynamics (MD) simulation, an in silico method, studies the motion of particles/atoms/molecules in a system over time. When they conducted an MD simulation study of VRC01 (a HIV-1 antibody) with gp120 (a viral envelope glycoprotein) and calculated the binding free energy, Zhang et al. [55] noticed that fluctuations of terminal residue E1 caused changes at the tip of the V5 loop in VRC01. Moreover, the D and V5 loops of VRC01 were subject to the largest conformational changes during binding with gp120, which accounts for neutralization resistance. The study has provided insight that will improve antibody design.

All-atom MD simulations of antibody SPE7 and its antigen were performed in an explicit solvent to investigate conformational changes and coupling during binding [56]. Quantitative analysis from MD trajectories and high-temperature unbinding kinetics analysis indicated that SPE7 follows a locally induced fit mechanism at the binding interface. This suggests that computational methods can be used to study recognition/interaction in antibody–antigen complexes.

Computer-aided Antibody Design

Epitope Predictions

Epitope identification is useful for epitope-based vaccine design and diagnostic development for a wide variety of diseases. The experimental identification of epitopes (using e.g., a phage display library, overlapping peptides, ELISA, immunofluorescence, immunohistochemistry, a radioimmunoassay, western blotting, X-ray crystallography, NMR studies of antibody–antigen structure, or attenuation of wild-type pathogens by random mutations and serial passages) is very costly and time-consuming. Therefore, predictive methods and software that focus on different types of epitopes have been used to predict epitopes of interest with decent accuracy (e.g., ABCpred [57], BepiPred [58], CEP [59], Discotope [60], EMT [61], Ellipro [62], FBCPred [63], LEPS [64], and PEPITO [65]).

The algorithms for epitope prediction tools are usually developed using machine learning techniques. These tools can be categorized as using “sequence-based” or “structure-based” prediction. Because there is an abundance of antibody–antigen complex sequences, sequence-based epitope prediction could yield higher accuracy. Structure-based epitope prediction tools would also be useful when an antigen or protein folds to form discontinuous epitopes. The structural changes that occur in an antibody when it binds to an antigen have also further complicated the epitope prediction calculation. Therefore, the combination of different algorithms balances prediction and overcomes the shortcoming of single constituent protocols. If the structure of an antigen has not been solved experimentally, protein structure prediction can aid in providing predictive structural conformations. Therefore, epitope prediction can be performed on both sequence- and structural-based algorithms for a more conclusive and accurate prediction. Epitope prediction modules can also be improved if pattern recognition is coupled with the antigenic propensity of amino acids (governed by e.g., polarity, flexibility, antigenicity, and surface accessibility) to provide additional data.

Recently, the structure of the 76-amino acid EPC1 protein from Echinococcus granulosus (the etiological agent responsible for cystic echinococcosis (CE) infection) was predicted by Etabar et al. [66] using a comparative modeling server [67]. Because the performance of EPC1 as an antigen in diagnostic assays might be affected by the presence of different isolates, the sequences of EPC1 obtained from different intermediate hosts of E. granulosus were assessed. The identity of the EPC1 sequence in the prevalent strains (G1 and G6 genotypes) showed that EPC1 is an antigenic protein and would be the appropriate candidate for serological assays when CE is prevalent in a region. In addition, nine B-cell epitope residues from the highly antigenic EPC1 protein have been identified from predictive structure via structural-based epitope prediction tools [60], which has further confirmed that EPC1 is an important antigenic protein.

In another study, Loyola et al. used structural bioinformatics tools to investigate the neuraminidase (NA) of Mexican A H1N1 influenza strain [68]. The NA protein sequences from the Mexican H1N1 virus (which caused a pandemic between April 2009 and October 2010) were aligned by structure-based sequence alignment (STRAP [69]) to identify NA-mutated and -conserved regions. The three-dimensional structures of wild- and mutant-type NA proteins were produced by comparative modeling (Modeller [70]). The structure was validated by a molecular docking simulation in which oseltamivir, a known NA ligand, was docked with the active site of the NA model. The docking results were reproducible when compared with the crystal structure [71]. Amino acids 448 to 469 were identified as an epitope from both sequence- and structure-based epitope predictors (ProPred [72]; MHC2Pred [73, 74]; CEP, Ellipro). A molecular docking simulation of the interaction between the NA protein and MHC II (by AutoDock 4.0.1 [75]) demonstrated that the identified epitope in the NA protein was able to dock in the binding pocket of MHC II. The interactions between the NA protein and MHC II were further investigated using MD simulation (by NAMD 2.6 [76]) and the data suggested that MHC II is able to recognize the selected epitope that is essential for antigen-specific immune responses. Antibody recognition assays conducted by the same authors have further validated theoretical calculation using bioinformatics, and have proved that both human and rabbit antibodies can recognize the epitope.

 Epitope Mapping of Antigen

Identification and characterization of antibody epitopes and antigen binding sites are crucial in antibody design for the development of therapeutics, vaccines, and diagnostics. Experimental epitope mapping by techniques such as array-based oligopeptide panning, mutagenesis mapping, phage display, and X-ray co-crystallography can be time consuming and costly. The challenges are more complicated when mapping complex target antigens such as membrane proteins or quaternary proteins.

Simonelli et al. investigated the structure of human anti-dengue antibody with dengue virus (DenV) serotypes [77]. They combined NMR epitope mapping to define the binding site of antibody DV32.6 against four DenV serotypes, and predicted the antibody–antigen complexes. The results provided adequate information for the rational design of antibody mutants with increased binding specificity and improved neutralization properties of up to 40-fold. Their NMR results validated the docking simulation, demonstrating that successful rational antibody design can be achieved, and the study of antibody–antigen interactions can be used to optimize antibody binding and other immunological properties.

The docking simulation for the homology modeling of antibody MoAb WH9 and the Der p 7 allergen has provided insight into the interaction between these proteins. Epitope mapping and in silico characterization revealed that residues L158 and D159 in MoAb WH9 are critical for the binding of the Der p 7 allergen [78]. The results have improved the understanding of the antigenicity of Der p 7 and its interactions with MoAb WH9 at the molecular level, and could facilitate the design of a better antibody.

Molecular dynamics (MD) simulation can also be applied to examine the conformational impact of mutation on antibody–antigen binding. In a study by Zhang et al. [79], MD simulation data revealed key residues in the framework regions that may have a structural impact on the CDRs; this was corroborated experimentally. Their computational work based on epitope scanning and MD simulation has established a novel method for antibody design.

 CDRs Design

For diagnostic in vitro applications, antibody engineering aims to modulate the antigen-binding affinity and specificity of the antibody for a more sensitive and specific immunoassay. At present, most antibodies are generated entirely using experimental approaches. Random mutagenesis and directed evolution protocols are only conducted to improve affinity after a promising antibody has been identified. The efficiency of experimental approaches would be improved if the position and type of a mutation were known. However, in computational modeling, computer-aided antibody design for diagnostic applications usually focuses on CDR regions, as well as the prediction of VL/VH domain orientation during antigen binding. Computer-aided antibody design could be used prior to antibody engineering to generate hypotheses for testing, thereby aiding experimental design and reducing the duration and cost of laboratory experiments.

Using a combination of a canonical structure for the CDRs of antibodies, refinement of the backbone structures of CDRs, and optimal amino acid selection for each position, Pantazes and Maranas demonstrated the de novo design of novel antibodies for a given antigen with promising antigen affinity [80]. They benchmarked the OptCDR algorithm with three test cases: the hapten fluorescein; a peptide from the capsid of the hepatitis C virus; and the vascular endothelial growth factor (VEGF) protein. The results demonstrated that OptCDR is able to retrieve novel sets of CDRs with comparable or even better interaction than the native antibodies, and features that correspond to high affinity binding specific to each of the three given antigens. In their later work, Pantazes and Maranas have also shown that antibody tertiary structures can be predicted by analyzing the structural features of affinity-matured antibodies [81]. Their database is able to predict antibody structures with an average RMSD of 1.9 ± 0.3 Å; their work has demonstrated that it is possible to quickly identify potential prototypes for any given antibody by incorporating the residue changes of the antibody relative to the prototype.

Yu et al. demonstrated the ability of computational methods to design CDRs [82]. They rationalized three-dimensional distributions to the interacting atoms from a protein structure database. Machine learning models were then generalized to design amino acid preferences in antibody–antigen interfaces. The test results indicated that structure-based computational antibody design can provide alternatives to experimental antibody design for predetermined antigen epitopes.

Choi and Deane used a database loop prediction technique to predict the structure of CDRs. The technique had an accuracy of less than 1.0 Å for all CDR loops except CDR-H3 with an RMSD of 2.3 Å [83]. Using both antigen-free and antigen-bound structures, the researchers were able to predict the bound structure of the hyper variable loop of CDR-H3 with an improved RMSD of 1.4 Å. They showed that the fragment-based search method was also able to predict CDR conformations accurately.

 Affinity Maturation of Antibody

In the process of understanding and predicting antibodies, aggregation is a common problem that is important in pharmaceutical development. Although loops engineering at Fab regions can improve antigen binding, they may also impair the conformational stability of an antibody. It may be desirable to disrupt the aggregation prone regions (APRs) during affinity maturation to avoid self-association.

A study on the potential aggregation liabilities from a human IgG1κ antibody was explored to mimic accelerated stability studies [84]. Two potential aggregation liabilities were identified in the VH domain of Fab from multiple elevated temperature MD simulation results. Data obtained from the study are useful for planning a rational structure-based strategy to design and select for specific and improved antibody candidates.

Schmidt et al. compared the binding properties of the wild-type common ancestor (UCA) and intermediate 2 (I-2) Fab with CH67 and CH65 [85]. They showed that antibody evolution in the B-cell lineage has preconfigured the CDR CH3 to yield a 30- to 40-fold increase in the association rate. Up to 5 μs of MD simulations on antibody–hemagglutinin (HA) complexes showed that antibody evolution in the lineage has reduced the flexibility of the early flexible CDRs of the heavy chains by two independent pathways. The results have improved our understanding and will aid the development of effective vaccines for influenza.

MD simulations of the Fc domains in human IgG1 have been used to study the effects of the single point mutation Q347E in CH3 domains [86]. Interaction and accessibility analyses have shown the potential effects of various mutations on fragment stability and binding characteristics. The Q347E mutation increased stability in the CH3 domain and affected interactions at the glycan residues. This work demonstrated that MD simulation supports the characterization and design of antibodies.

Limitations & Challenges

Structural bioinformatics is an established field in biology and supports experimental results in various ways. The advantages of theoretical over experimental results are clear: mathematical and computational calculations are cheaper and quicker, and parameter scales can be tuned to study the influence of different environmental parameters. The cliché of structural bioinformatics is that predictive models are only as good as the available data and the theory on which they are based. Every result obtained from a theoretical calculation has to be verified experimentally. Some parameters have to be estimated because they are not known and are not accessible from experiments.

Continuous improvement in computational resources has made structural bioinformatics an indispensable tool in biomedical research. The main challenge from a diagnostic perspective is the accuracy of protein structure prediction, be it the structure of an antibody or an antigen. When the structure of a protein cannot be predicted using a modeling technique with a high level of confidence, i.e., comparative modeling, ab initio approaches with improved force field accuracy are required, but they limit the resulting energy landscape.

Regarding antibody–antigen recognition, the docking process needs to take into account the flexibility, solvation, and entropy during the binding of the antigen to the antibody. Approximations in docking scoring functions to estimate the binding free energy are a challenge, as is the fact that complexes are modeled with united-atom or coarse-grained methods in implicit solvents. Improvements to force field accuracy and speed as well as conformational entropy solvent polarization are also challenging.

Relatively small errors in each step of a computational calculation can be amplified in subsequent steps. The complexity of antibody–antigen interactions make predictions more complicated unless the proper structure of the complex is solved. However, protein engineering and advances in computer science will improve such research.

 Future Perspectives

Structural bioinformatics can be applied to corroborate or challenge a theory, or to search for optimal conditions for a desired outcome. The choice of modeling techniques depends on the question to be answered. Researchers can either use a single-method approach or a combination of approaches depending on their analysis requirements. With the advances in our understanding and improvements in techniques, computational calculation will one day produce more accurate prediction. This will facilitate its application to the early stages of antibody–antigen design to improve binding affinity and specificity, which will reduce the time and cost of producing an in vitro antigen-based diagnostic test.


We would like to thank the Malaysian Ministry of Education for the following grants: HICoE (311/CIPPM/44001005) and ERGS (203/CIPPM/44001005).


  1. Epstein CJ, Goldberger RF, Anfinsen CB (1963) The genetic control of tertiary protein structure: studies with model systems. Cold Spring Harbor Symp Quant Biol, 28:439-449.
  2. Chothia C, Lesk AM (1986) The relation between the divergence of sequence and structure in proteins. EMBO 5(4):823-826.
  3. Rost B (1999) Twilight zone of protein sequence alignments. Protein Eng 12(2):85-94.
  4. Sander C, Schneider R: (1991) Database of homology-derived protein structures and the structural meaning of sequence alignment. Proteins 9(1):56-68.
  5. Marti-Renom MA, Stuart AC, Fiser A, Sanchez R, Melo F, Sali A (2000) Comparative protein structure modeling of genes and genomes. Annu Rev Biophys Biomol Struct, 29:291-325.
  6. Tramontano A, Cozzetto D, Giorgetti A, Raimondo D: The assessment of methods for protein structure prediction. Methods Mol Biol 2008, 413:43-57.
  7. DeWeese-Scott C, Moult J: Molecular modeling of protein function regions. Proteins 2004, 55(4):942-961.
  8. Carlsson J, Coleman RG, Setola V, Irwin JJ, Fan H, Schlessinger A, Sali A, Roth BL, Shoichet BK: Ligand discovery from a dopamine D3 receptor homology model and crystal structure. Nat Chem Biol 2011, 7(11):769-778.
  9. Schlessinger A, Geier E, Fan H, Irwin JJ, Shoichet BK, Giacomini KM, Sali A: Structure-based discovery of prescription drugs that interact with the norepinephrine transporter, NET. Proc Natl Acad Sci USA 2011, 108(38):15810-15815.
  10. Pieper U, Schlessinger A, Kloppmann E, Chang GA, Chou JJ, Dumont ME, Fox BG, Fromme P, Hendrickson WA, Malkowski MG et al: Coordinating the impact of structural genomics on the human alpha-helical transmembrane proteome. Nat Struct Mol Biol 2013, 20(2):135-138.
  11. Anfinsen CB: Principles that govern the folding of protein chains. Science 1973, 181(4096):223-230.
  12. Anfinsen CB, Haber E, Sela M, White FWJ: The kinetics of formation of native ribonuclease during oxidation of the reduced polypeptide chain. Proc Natl Acad Sci USA 1961, 47:1309-1314.
  13. Chowdhury S, Lee MC, Xiong G, Duan Y: Ab initio folding simulation of the Trp-cage mini-protein approaches NMR resolution. J Mol Biol 2003, 327(3):711-717.
  14. Pitera JW, Swope W: Understanding folding and design: replica-exchange simulations of "Trp-cage" miniproteins. Proc Natl Acad Sci USA 2003, 100(13):7587-9752.
  15. Simmerling C, Strockbine B, Roitberg AE: All-atom structure prediction and folding simulations of a stable protein. J Am Chem Soc 2002, 124(38):11258-11259.
  16. Ben-David M, Noivirt-Brik O, Paz A, Prilusky J, Sussman JL, Levy Y: Assessment of CASP8 structure predictions for template free targets. Proteins 2009, 77 Suppl 9:50-65.
  17. Bradley P, Misura KM, Baker D: Toward high-resolution de novo structure prediction for small proteins. Science 2005, 309(5742):1868-1871.
  18. Kinch L, Yong Shi S, Cong Q, Cheng H, Liao Y, Grishin NV: CASP9 assessment of free modeling target predictions. Proteins 2011, 79 Suppl 10:59-73.
  19. Klepeis JL, Wei Y, Hecht MH, Floudas CA: Ab initio prediction of the three-dimensional structure of a de novo designed protein: a double-blind case study. Proteins 2005, 58(3):560-570.
  20. Wu S, Skolnick J, Zhang Y: Ab initio modeling of small proteins by iterative TASSER simulations. BMC Biol 2007, 5:17.
  21. Xu D, Zhang Y: Ab initio protein structure assembly using continuous structure fragments and optimized knowledge-based force field. Proteins 2012, 80(7):1715-1735.
  22. Zhang Y, Kolinski A, Skolnick J: TOUCHSTONE II: a new approach to ab initio protein structure prediction. Biophys J 2003, 85(2):1145-1164.
  23. Manning T, Sleator RD, Walsh P: Biologically inspired intelligent decision making: A commentary on the use of artificial neural networks in bioinformatics. Bioengineered 2014, 5:4-19.
  24. Vanhala J, Kaski K: Protein structure prediction system based on artificial neural networks. Proc Int Conf Intell Syst Mol Biol 1993, 1:402-410.
  25. Bystroff C, Krogh A: Hidden Markov Models for prediction of protein features. Methods Mol Biol 2008, 413:173-198.
  26. Soding J: Protein homology detection by HMM-HMM comparison. Bioinformatics 2005, 21(7):951-960.
  27. Chatterjee P, Basu S, Kundu M, Nasipuri M, Plewczynski D: PSP_MCSVM: brainstorming consensus prediction of protein secondary structures using two-stage multiclass support vector machines. J Mol Model 2011, 17(9):2191-2201.
  28. Chen B, Johnson M: Protein local 3D structure prediction by Super Granule Support Vector Machines (Super GSVM). BMC Bioinformatics 2009, 10 Suppl 11:S15.
  29. He J, Hu HJ, Harrison R, Tai PC, Pan Y: Rule generation for protein secondary structure prediction with support vector machines and decision tree. IEEE Trans Nanobioscience 2006, 5(1):46-53.
  30. Kountouris P, Hirst JD: Prediction of backbone dihedral angles and protein secondary structure using support vector machines. BMC Bioinformatics 2009, 10:437.
  31. Bouziane H, Messabih B, Chouarfia A: Profiles and majority voting-based ensemble method for protein secondary structure prediction. Evol Bioinform Online 2011, 7:171-189.
  32. Zou L, Wang Z, Wang Y, Hu F: Combined prediction of transmembrane topology and signal peptide of beta-barrel proteins: using a hidden Markov model and genetic algorithms. Comput Biol Med 2010, 40(7):621-628.
  33. Rorsman P, Berggren PO, Bokvist K, Ericson H, Mohler H, Ostenson CG, Smith PA: Glucose-inhibition of glucagon secretion involves activation of GABAA-receptor chloride channels. Nature 1989, 341(6239):233-236.
  34. Tong JC, Myers MA, Mackay IR, Zimmet PZ, Rowley MJ: The PEVKEK region of the pyridoxal phosphate binding domain of GAD65 expresses a dominant B cell epitope for type 1 diabetes sera. Ann N Y Acad Sci 2002, 958:182-189.
  35. Tree TI, Morgenthaler NG, Duhindan N, Hicks KE, Madec AM, Scherbaum WA, Banga JP: Two amino acids in glutamic acid decarboxylase act in concert for maintenance of conformational determinants recognised by Type I diabetic autoantibodies. Diabetologia 2000, 43(7):881-889.
  36. Wiltgen M, Tilz GP: Homology modelling: a review about the method on hand of the diabetic antigen GAD 65 structure prediction. Wien Med Wochenschr 2009, 159(5-6):112-215.
  37. Carman WF, Trautwein C, van Deursen FJ, Colman K, Dornan E, McIntyre G, Waters J, Kliem V, Muller R, Thomas HC et al: Hepatitis B virus envelope variation after transplantation with and without hepatitis B immune globulin prophylaxis. Hepatology 1996, 24(3):489-493.
  38. Carman WF, Van Deursen FJ, Mimms LT, Hardie D, Coppola R, Decker R, Sanders R: The prevalence of surface antigen variants of hepatitis B virus in Papua New Guinea, South Africa, and Sardinia. Hepatology 1997, 26(6):1658-1666.
  39. Waters JA, Kennedy M, Voet P, Hauser P, Petre J, Carman W, Thomas HC: Loss of the common "A" determinant of hepatitis B surface antigen by a vaccine-induced escape mutant. J Clin Invest 1992, 90(6):2543-2547.
  40. Carman WF, Zanetti AR, Karayiannis P, Waters J, Manzillo G, Tanzi E, Zuckerman AJ, Thomas HC: Vaccine-induced escape mutant of hepatitis B virus. Lancet 1990, 336(8711):325-329.
  41. Chen YC, Delbrook K, Dealwis C, Mimms L, Mushahwar IK, Mandecki W: Discontinuous epitopes of hepatitis B surface antigen derived from a filamentous phage peptide library. Proc Natl Acad Sci USA 1996, 93(5):1997-2001.
  42. Ie SI, Thedja MD, Roni M, Muljono DH: Prediction of conformational changes by single mutation in the hepatitis B virus surface antigen (HBsAg) identified in HBsAg-negative blood donors. Virol J 2010, 7:326.
  43. Ranjbar MM, Ghorban K, Alavian SM, Keyvani H, Dadmanesh M, Roayaei Ardakany A, Motedayen MH, Sazmand A: GB virus C/hepatitis G virus envelope glycoprotein E2: Computational molecular features and immunoinformatics study. Hepat Mon 2013, 13(12):e15342.
  44. Choong YS, Lim TS, Chew AL, Aziah I, Ismail A: Structural and functional studies of a 50 kDa antigenic protein from Salmonella enterica serovar Typhi. J Mol Graph Model 2011, 29(6):834-842.
  45. Yung-Hung RL, Ismail A, Lim TS, Choong YS: A 35 kDa antigenic protein from Shigella flexneri: in silico structural and functional studies. Biochem Biophys Res Commun 2011, 415(2):229-234.
  46. Rahbar MR, Rasooli I, Gargari SL, Sandstrom G, Amani J, Fattahian Y, Jahangiri A, Jalali M: A potential in silico antibody-antigen based diagnostic test for precise identification of Acinetobacter baumannii. J Theor Biol 2012, 294:29-39.
  47. Ewert S, Honegger A, Pluckthun A: Stability improvement of antibodies for extracellular and intracellular applications: CDR grafting to stable frameworks and structure-based framework engineering. Methods 2004, 34(2):184-199.
  48. Honegger A, Malebranche AD, Rothlisberger D, Pluckthun A: The influence of the framework core residues on the biophysical properties of immunoglobulin heavy chain variable domains. Protein Eng Des Sel 2009, 22(3):121-134.
  49. Lee VS, Tue-ngeun P, Nangola S, Kitidee K, Jitonnom J, Nimmanpipug P, Jiranusornkul S, Tayapiwatana C: Pairwise decomposition of residue interaction energies of single chain Fv with HIV-1 p17 epitope variants. Mol Immunol 2010, 47(5):982-990.
  50. Tue-ngeun P, Kodchakorn K, Nimmanpipug P, Lawan N, Nangola S, Tayapiwatana C, Rahman NA, Zain SM, Lee VS: Improved scFv anti-HIV-1 p17 binding affinity guided from the theoretical calculation of pairwise decomposition energies and computational alanine scanning. Biomed Res Int 2013, 2013:713585.
  51. Narayanan A, Sellers BD, Jacobson MP: Energy-based analysis and prediction of the orientation between light- and heavy-chain antibody variable domains. J Mol Biol 2009, 388(5):941-953.
  52. Sela-Culang I, Alon S, Ofran Y: A systematic comparison of free and bound antibodies reveals binding-related conformational changes. J Immunol 2012, 189(10):4890-4899.
  53. Corrada D, Morra G, Colombo G: Investigating allostery in molecular recognition: insights from a computational study of multiple antibody-antigen complexes. J Phys Chem B 2013, 117(2):535-552.
  54. Sircar A, Gray JJ: SnugDock: paratope structural optimization during antibody-antigen docking compensates for errors in antibody homology models. PLoS Comput Biol 2010, 6(1):e1000644.
  55. Zhang Y, Pan D, Shen Y, Jin N, Liu H, Yao X: Understanding the molecular mechanism of the broad and potent neutralization of HIV-1 by antibody VRC01 from the perspective of molecular dynamics simulation and binding free energy calculations. J Mol Model 2012, 18(9):4517-4527.
  56. Wang W, Ye W, Yu Q, Jiang C, Zhang J, Luo R, Chen HF: Conformational selection and induced fit in specific antibody and antigen recognition: SPE7 as a case study. J Phys Chem B 2013, 117(17):4912-4923.
  57. Saha S, Raghava GP: Prediction of continuous B-cell epitopes in an antigen using recurrent neural network. Proteins 2006, 65(1):40-48.
  58. Larsen JE, Lund O, Nielsen M: Improved method for predicting linear B-cell epitopes. Immunome Res 2006, 2:2-5.
  59. Kulkarni-Kale U, Bhosle S, Kolaskar AS: CEP: a conformational epitope prediction server. Nucleic Acids Res 2005, 33(Web Server issue):W168-171.
  60. Andersen PH, Nielsen M, Lund O: Prediction of residues in discontinuous B-cell epitopes using protein 3D structures. Protein Sci 2006, 15(11):2558-2567.
  61. Batori V, Friis EP, Nielsen H, Roggen EL: An in silico method using an epitope motif database for predicting the location of antigenic determinants on proteins in a structural context. J Mol Recognit 2006, 19(1):21-29.
  62. Ponomarenko J, Bui HH, Li W, Fusseder N, Bourne PE, Sette A, Peters B: ElliPro: a new structure-based tool for the prediction of antibody epitopes. BMC Bioinformatics 2008, 9:514.
  63. El-Manzalawy Y, Dobbs D, Honavar V: Predicting linear B-cell epitopes using string kernels. J Mol Recognit 2008, 21(4):243-255.
  64. Wang HW, Lin YC, Pai TW, Chang HT: Prediction of B-cell linear epitopes with a combination of support vector machine classification and amino acid propensity identification. J Biomed Biotechnol 2011, 2011:432830.
  65. Sweredoski MJ, Baldi P: PEPITO: improved discontinuous B-cell epitope prediction using multiple distance thresholds and half sphere exposure. Bioinformatics 2008, 24(12):1459-1460.
  66. Etebar F, Jalousian F, Hosseini SH, Kordafshari S, Najafi A: Immunoproteomics approach for EPC1 antigenic epitope prediction of G1 and G6 strains of Echinococcus granulosus. Parasitol Res 2013, 112(9):3129-3135.
  67. Arnold K, Bordoli L, Kopp J, Schwede T: The SWISS-MODEL workspace: a web-based environment for protein structure homology modelling. Bioinformatics 2006, 22(2):195-201.
  68. Loyola PK, Campos-Rodriguez R, Bello M, Rojas-Hernandez S, Zimic M, Quiliano M, Briz V, Munoz-Fernandez MA, Tolentino-Lopez L, Correa-Basurto J: Theoretical analysis of the neuraminidase epitope of the Mexican A H1N1 influenza strain, and experimental studies on its interaction with rabbit and human hosts. Immunol Res 201, 56(1):44-60.
  69. Gille C, Frommel C: STRAP: editor for STRuctural Alignments of Proteins. Bioinformatics 2001, 17(4):377-378.
  70. Sali A, Blundell TL: Comparative protein modelling by satisfaction of spatial restraints. J Mol Biol 1993, 234(3):779-815.
  71. Collins PJ, Haire LF, Lin YP, Liu J, Russell RJ, Walker PA, Skehel JJ, Martin SR, Hay AJ, Gamblin SJ: Crystal structures of oseltamivir-resistant influenza virus neuraminidase mutants. Nature 2008, 453(7199):1258-1261.
  72. Singh H, Raghava GP: ProPred: prediction of HLA-DR binding sites. Bioinformatics 2001, 17(12):1236-1237.
  73. Bhasin M, Raghava GP: SVM based method for predicting HLA-DRB1*0401 binding peptides in an antigen sequence. Bioinformatics 2004, 20(3):421-423.
  74. Lata S, Bhasin M, Raghava GP: Application of machine learning techniques in predicting MHC binders. Methods Mol Biol 2007, 409:201-215.
  75. Morris GM, Goodsell DS, Halliday RS, Huey R, Hart WE, Belew RK, Olson AJ: Automated docking using a Lamarckian genetic algorithm and an empirical binding free energy function. J Comput Chem 1998, 19(14):1639-1662.
  76. Phillips JC, Braun R, Wang W, Gumbart J, Tajkhorshid E, Villa E, Chipot C, Skeel RD, Kale L, Schulten K: Scalable molecular dynamics with NAMD. J Comput Chem 2005, 26(16):1781-1802.
  77. Simonelli L, Pedotti M, Beltramello M, Livoti E, Calzolai L, Sallusto F, Lanzavecchia A, Varani L: Rational engineering of a human anti-dengue antibody through experimentally validated computational docking. PLoS One 2013, 8(2):e55561.
  78. Tai HY, Zhou JK, Chou H, Tam MF, Chen YS, Sheu SY, Shen HD: Epitope mapping and in silico characterization of interactions between Der p 7 allergen and MoAb WH9. PLoS One 2013, 8(8):e71269.
  79. Zhang D, Chen CF, Zhao BB, Gong LL, Jin WJ, Liu JJ, Wang JF, Wang TT, Yuan XH, He YW: A novel antibody humanization method based on epitopes scanning and molecular dynamics simulation. PLoS One 2013, 8(11):e80636.
  80. Pantazes RJ, Maranas CD: OptCDR: a general computational method for the design of antibody complementarity determining regions for targeted epitope binding. Protein Eng Des Sel 2010, 23(11):849-858.
  81. Pantazes RJ, Maranas CD: MAPs: a database of modular antibody parts for predicting tertiary structures and designing affinity matured antibodies. BMC Bioinformatics 2013, 14(1):168.
  82. Yu CM, Peng HP, Chen IC, Lee YC, Chen JB, Tsai KC, Chen CT, Chang JY, Yang EW, Hsu PC et al: Rationalization and design of the complementarity determining region sequences in an antibody-antigen recognition interface. PLoS One 2012, 7(3):e33340.
  83. Choi Y, Deane CM: Predicting antibody complementarity determining region structures without classification. Mol Biosyst 2011, 7(12):3327-3334.
  84. Buck PM, Kumar S, Singh SK: Insights into the potential aggregation liabilities of the b12 Fab fragment via elevated temperature molecular dynamics. Protein Eng Des Sel 2012, 26(3):195-205.
  85. Schmidt AG, Xu H, Khan AR, O'Donnell T, Khurana S, King LR, Manischewitz J, Golding H, Suphaphiphat P, Carfi A et al: Preconfiguration of the antigen-binding site during affinity maturation of a broadly neutralizing influenza virus antibody. Proc Natl Acad Sci USA 2013, 110(1):264-269.
  86. Lai B, Hasenhindl C, Obinger C, Oostenbrink C: Molecular dynamics simulation of the crystallizable fragment of IgG1-Insights for the design of Fcabs. Int J Mol Sci 2014, 15(1):438-455.