Prediction of mutation hotspots in antigens

Abstract

The present invention relates to a computer implementation method for predicting site-directed mutagenesis of an antigen, the method comprising: (a) providing in silico at least one three-dimensional structure of the antigen, preferably an ensemble of three-dimensional structures of the antigen; (b) providing in silico at least one three-dimensional structure of at least one antibody or fragment thereof that is binding to or specific to the antigen, preferably an ensemble of three-dimensional structures of the antibody or fragment thereof; and (c) determining in silico the accessibility of the structure of (a) to the structure of (b), indicating that higher accessibility indicates higher mutagenesis, particularly under the selective pressure exerted by the antibody or fragment thereof, thereby predicting the site-directed mutagenesis.

Description

[Technical Field]

[0001] The present invention relates to a computer implementation method for predicting site-directed mutagenesis of an antigen, the method comprising: (a) providing in silico at least one three-dimensional structure of the antigen, preferably an ensemble of three-dimensional structures of the antigen; (b) providing in silico at least one three-dimensional structure of at least one antibody or fragment thereof that is binding to or specific to the antigen, preferably an ensemble of three-dimensional structures of the antibody or fragment thereof; and (c) determining in silico the accessibility of the structure of (a) to the structure of (b), wherein higher accessibility indicates higher mutagenesis, particularly under the selective pressure exerted by the antibody or fragment thereof, thereby predicting the site-directed mutagenesis. [Background technology]

[0002] This specification references numerous documents, including patent applications and manufacturers' manuals. While the disclosures of these documents are not considered relevant to the patentability of the present invention, they are incorporated herein by reference in their entirety. More specifically, all referenced documents are incorporated by reference to the same extent as any individual document is specifically and individually indicated to be incorporated by reference.

[0003] The immune system is generally capable of recognizing pathogens and novel variants of pathogens and triggering a response. Biotechnological means that stimulate the immune system, such as vaccines, and therapeutic drugs based on components of the immune system, such as therapeutic antibodies, are constantly being developed and improved. However, insufficient immune responses and inadequate stimulating effects of vaccines have been, and continue to be, difficult challenges to overcome. This is well known, for example, in recurrent influenza infections and has also been evident in the COVID-19 pandemic.

[0004] Pathogens such as viruses and bacteria are constantly exploring the effects of genetic mutations on their phenotypes. This makes the development of vaccines for infections that have not yet manifested or for variants that have not yet appeared difficult. Furthermore, both the immune system and vaccines and therapeutic antibodies exert selective pressure, leading to the dominance of pathogen variants that particularly evade the immune system. This phenomenon is also known as immune evasion.

[0005] The COVID-19 pandemic, caused by severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2), has entered a new phase with the emergence of the Omicron variant (BA.1-BA.5). Of several past and present variants (VoCs) of concern, Omicron BA.1 is the first evasive variant, and immunity conferred by infection with previous variants or vaccination has been significantly reduced. The induction of cross-neutralizing antibodies against Omicron infection from serum collected from patients already infected with previous VoCs has been strongly reduced. The reduced vaccine efficacy is mitigated by additional vaccinations ("boosters").

[0006] The spike (S) protein is a homotrimeric glycoprotein exposed on the viral surface and is the primary target of either innate or stimulated immune responses, thus being subjected to maximum immunosuppression. The Omicron BA.1 spike protein has 39 changes, counting mutations, deletions, and insertions of amino acids. Many of these changes are in the receptor-binding domain (RBD), which in the SARS-CoV-2 spike protein is found in either a closed (i.e., nearly protected from the immune system) or open conformation. In the open conformation, it can bind to the human angiotensin-converting enzyme 2 (ACE2) receptor, thereby mediating attachment to and entry into cells. Omicron subsequently evolved further through BA.2 to BA.4 / BA.5, acquiring 11 additional mutations and losing 14 mutations compared to BA.1.

[0007] Elucidating the factors that cause spike protein mutations and the ability to predict new mutation sites in future variants are of great interest, given the impact of newly emerging variants on the trajectory of pandemics. Deep mutational scanning results showed that the sublineages BA.2.12.1, BA.4, and BA.5 exhibit neutralization evasion to both vaccination (using wild-type spike) and BA.1 infection (Cao et al., Nature 602(7898):657-663, 2022). This suggests that Omicron continues to evolve to evade responses to previously circulating variants. In a recent study, deep mutational learning techniques were used to evaluate the impact of RBD mutations on RBD-ACE2 binding from a yeast screening library (Taft et al., Cell doi:10.1016 / j.cell.2022.08.024, 2022). The authors predicted the range of future antibody-evading RBD variants. Previously, Bai et al. calculated the binding free energy of spike-antibody and ACE2 receptor complexes based on a coarse-grained model to predict the effects of spike mutations (Bai et al., Journal of the American Chemical Society 143(42):17646-17654, 2021). Chen et al. trained a classification network using short molecular dynamics (MD) simulations of spike variants to predict changes in the binding free energy of RBD-ACE2 interactions (Chen et al., Proceedings of the National Academy of Sciences USA 118(42):e2106480118, 2021). Starr et al. systematically investigated the effects of RBD mutations on protein expression, ACE2 interaction, and antibody recognition using deep mutation scanning (Starr et al., Cell 182(5):1295-1310.e20, 2020; Starr et al., Science 371(6531):850-854, 2021).Recently, structural studies have revealed that the RBD-ACE2 binding affinity of the omicron variant is similar to that of the delta variant, and that the omicron spike exhibits significant antibody evasion (Mannar et al., Science 375(6582):766-764, 2022).

[0008] In a prior study co-authored by the present inventors (Sikora et al., PLoS Comput Biol 17(4):e1008790, 2021), a computational epitope map of the SARS-CoV-2 spike protein is presented. This study uses a composite scoring scheme based on accessibility, rigidity, conservation, and immunogenicity scores. This study does not mention the prediction of mutation hotspots or immune evasion. Surprisingly, the present inventors have found that a scoring scheme that is conceptually simpler than the scoring scheme of Sikora et al., op. cit., is superior to the aforementioned composite scoring scheme in terms of identifying and predicting mutation hotspots. [Overview of the project]

[0009] Considering the prior art and its shortcomings, the underlying technical problem of the present invention can be seen as providing improved means and methods for predicting mutational hotspots in antigens, including antigens of variants of pathogens of concern as potentially causing a pandemic.

[0010] This technical problem is solved by the subject matter of the claims, as demonstrated by the examples. Therefore, in a first embodiment, the present invention provides a computer-implemented method for predicting site-directed mutagenicity of an antigen, the method being: (a) Providing in silico at least one three-dimensional structure of the antigen, preferably an ensemble of three-dimensional structures of the antigen, (b) Providing in silico at least one three-dimensional structure of at least one antibody or fragment thereof that is bindable to or specific to the antigen, preferably an ensemble of three-dimensional structures of the antibody or fragment thereof, (c) The accessibility of structure (a) to structure (b) is determined in silico, and the higher the accessibility, the higher the mutagenic activity, especially under the selective pressure exerted by the antibody or its fragment, Includes, This allows us to predict the site-directed mutagenesis activity.

[0011] The term "antigen" has an established meaning in this field. An antigen refers to a molecule that can trigger an immune response, such as polypeptides, proteins, post-translational modifications like polysaccharides, and combinations thereof. Antigens are recognized and bound, in particular, by antibodies or fragments thereof. Antigens, especially those of pathogens, are known to evolve. Due to the limited precision of the mechanisms for replicating genetic material, the search for new variants is constant. Variants that are advantageous under given circumstances generally become dominant. Of interest here are the situations in which the mammalian immune system exerts selective pressure. Factors that exert such pressure include, for example, antibodies that are enhanced against pathogens, such as antibodies produced in response to vaccination.

[0012] For vaccine design purposes, it is of interest to know which specific regions or epitopes of antigens encoded by a pathogen are susceptible to mutation. This is because such mutations can induce immune evasion, where a previously effective vaccine fails to elicit a satisfactory response to a new variant of the pathogen. This invention focuses on identifying or predicting regions where increased mutagenicity is expected, particularly under the selective pressure of components of the mammalian immune system. In other words, this invention makes it possible to assess the suitability of a given antigen in terms of its ability to evade antibody-based immune responses. Such suitability information can generally be obtained at a high structural resolution at the amino acid level. Notably, as demonstrated in the examples, the method of this invention has successfully predicted a new variant of SARS-CoV-2. This was achieved using only the limited structural and genetic information of the virus available at the time of the pandemic. Such information is invaluable when designing and developing countermeasures against pathogens, especially those that can cause serious harm and / or potentially pandemic pathogens. Often, such countermeasures relate to the immune system, such as vaccines, therapeutic vaccines, and therapeutic antibodies.

[0013] For this purpose, in silico representations of biomolecular structures, particularly polypeptide and protein structures, are employed. Such representations are datasets that can be stored in computing or data storage devices. For example, such a representation may be a list of three-dimensional coordinates of the constituent atoms and / or groups of atoms of the molecule of interest. The coordinates may be Cartesian coordinates, but are not required. Any coordinate system that defines three-dimensional space, such as polar coordinates, can be used. Decades of effort in structure determination and prediction have made many three-dimensional structural representations available and can also be generated. For databases and methods, see below.

[0014] In in silico structures, we can distinguish between representations that provide a single representative set of coordinates corresponding to a single structure, and datasets that provide multiple coordinate sets for a given molecule. The latter is also called a structural ensemble. Of particular interest here is the structural ensemble that represents the conformational space available to a given molecule. It is well established that proteins and polypeptides are flexible, so a single structure may be representative, but information about flexibility is lacking. This method can also be implemented with such a single representative structure, but we prioritize structural ensembles.

[0015] As will be disclosed in detail below, an ensemble of structures can be generated from a single starting structure, or information about the flexibility contained in the starting ensemble of structures can be further enriched, thereby transforming such a starting ensemble of structures into a final ensemble of structures, the latter containing more structures and / or more complete information about the conformational space available to a given molecule. In either case, methods of structural exploration can be employed. A preferred method of structural exploration is molecular dynamics simulation (MD). Since the results of structural determination by NMR are generally expressed as an ensemble, an example of a starting ensemble of structures is an ensemble of NMR structures.

[0016] The term "antibody" has an established meaning in the relevant field and generally refers to an element of the humoral immune response that the immune system can produce against an antigen, for example, to fight infection. In addition to antibodies formed by the immune system, antibody fragments, particularly those containing antigen-binding sites or complementarity-determining regions (CDRs), can also be employed. Such fragments are well known and include, for example, Fab, F(ab')2, scFv, dsFv, ds-scFv, and Diabody.

[0017] Antibodies and their fragments can, appropriately, bind to a given antigen. Preferably, such binding is specific, i.e., the affinity of an antibody or its fragment for a given antigen (homologous antigen) is higher than for other possible antigens.

[0018] Step (c) of the method of the first aspect provides for determining the mutual accessibility of the structure of (a) and the structure of (b), preferably the mutual accessibility of the constituent building blocks (constituents) of the molecule under consideration. In other words, what is determined is the accessibility of the antigen to the antibody or its fragment, which is also abbreviated herein as AA (antibody accessibility), and the score determined by the method of the present invention is also referred to as the antibody accessibility score (AAS).

[0019] In a preferred embodiment, the accessibility of the amino acid residues of the antigen and the amino acid residues of the antibody or its fragment is evaluated. On the other hand, post-translational modifications (PTMs) may be considered if desired. See the following for details.

[0020] Generally, accessibility is a measure of proximity. In other words, for the structure of a given antigen and the structure of a given antibody or the structure of a given fragment, if a predefined proximity between an amino acid of the antigen and an amino acid of either the antibody or the fragment is possible in three-dimensional space, the amino acid of the antigen is considered accessible to the antibody or the fragment. As will be described in more detail below, it is envisioned to zoom in on specific parts of the antibody or antigen. For example, within an antibody or its fragment, the CDRs can be considered individually so as to determine whether a given amino acid of the antigen is accessible to the CDR.

[0021] Accessibility can be evaluated at various levels depending on whether the structural representation for this purpose is at a detailed or coarse level. For example, accessibility (and the underlying proximity) may be evaluated at the atomic level. On the other hand, at the level of larger building blocks, in the case of polypeptides and proteins, it may be evaluated at the amino acid level. Further details regarding this point will be further disclosed below.

[0022] Accessibility can be influenced by several factors, and considering these is the key to the present invention. These factors include the flexibility of the molecule (antigen, antibody or its fragment), non-amino acid building blocks, particularly the presence of post-translational modifications.

[0023] As a result of flexibility, various conformations are available for a single molecule. A given amino acid or atom of an antigen may be accessible to an antibody or its fragment (fragment) in one conformation, but not in a different conformation of the same antigen, the same antibody or its fragment. The accessibility according to the present invention can be evaluated based on a single structure, but preference is given to using an ensemble of structures. When using an ensemble of structures for the purpose of determining accessibility, one method is to determine the accessibility for each possible combination of structures from any of the ensembles. For example, if the antigen structure ensemble contains n members (parts) and the antibody or fragment structure ensemble contains m members, n×m combinations are possible, and for each of these combinations, the accessibility between each amino acid or atom of the antigen and each amino acid or atom of the antibody can be determined.

[0024] While we do not wish to be bound by any particular theory, an ensemble can be considered to represent the conformational space available to a given molecule, and using an ensemble to determine accessibility provides a more realistic picture of real-world accessibility. Accessibility, while we do not wish to be bound by any particular theory, is thought to correlate with the likelihood of a given region, site, epitope, or amino acid of an antigen changing or mutating. This is because the more accessible an amino acid or epitope is to an antibody, the more likely the pathogen is to benefit from a mutation in that amino acid or epitope, and the more likely the recognition by that antibody will be impaired.

[0025] Mutations alter the amino acids of an antigen, and the CDR of an antibody is formed by amino acids; therefore, the methods of the present invention preferably provide a determination of interoperability at the amino acid residue level. However, proteins and polypeptides may contain, and often do contain, components other than amino acids. After synthesis on ribosomes, polypeptides or proteins often undergo post-translational modifications (PTMs). While we do not wish to be bound by any particular theory, it is thought that the PTMs contribute less to antigenicity than the amino acids of the antigen, and also less to the ability of the antibody or fragment to recognize the antigen. The expression "less contribution" reflects a stepwise phenomenon. Some PTMs may contribute to antigenicity, others less, and still others may not contribute at all. In one approach (which was very successful; see examples), PTMs were not considered at all when determining the interoperability of structures. However, the present invention is not limited in that way.

[0026] In other words, in a preferred embodiment, the present invention can be carried out using antigens and antibodies based solely on amino acid composition. This may be particularly applicable when PTMs do not occur or when there is no knowledge of PTMs. However, it is preferred to appropriately consider PTMs, whether on the antigen or on the antibody or fragment. In a preferred embodiment, PTMs can be considered as additions to the antibody and / or antigen in each ensemble. Preferably, each member of the ensemble of antibody or fragment and / or antigen structures containing PTMs can be expanded by considering multiple combinations and conformations of the PTMs.

[0027] From the above, it can be concluded that any PTM on a molecule may provide a shielding effect against interactions (amino acid-amino acid interactions) that are thought to be related to immunity or immune evasion. In other words, shielded residues are less likely to become mutation hotspots, either in a single structure or over a long period of time. "Shielded" is equivalent to being partially inaccessible or only partially accessible, and the concept of partial accessibility refers to the implementation of the present invention that employs an ensemble of structures and / or considers multiple PTMs for each structure. A given residue may be accessible in some structures included in the ensemble under consideration, but inaccessible in some other structures.

[0028] While it is preferable to use an ensemble representation of each structure as a whole, it is also preferable to use a single structure for each amino acid chain and sample the conformational space available for PTM, in order to capture the above-mentioned shielding effect. See, for example, doi.org / 10.1101 / 2021.08.04.455134.

[0029] Surprisingly, the present invention, which preferably uses only the mutual accessibility between antigen and antibody or fragment, surpasses prior art methods that use solvent accessibility or more complex composite scoring schemes. Indeed, it is unexpected that a measure that reflects purely physical accessibility from a steric perspective (accessibility to homogeneous molecules rather than to water) yields the performance demonstrated by the examples.

[0030] In other words, even the concept of accessibility as defined herein is sufficient to achieve excellent performance. However, further advantages of the present invention become apparent when PTMs are present and considered when determining accessibility. As described above, the handling of PTMs may be adjusted, preferably based on existing knowledge of whether a given PTM is known or expected to contribute to antigenicity, or whether its primary role is rather the shielding of structural elements, particularly amino acid residues, that contribute to antigenicity. As previously stated, in one preferred embodiment, PTM-PTM contact and PTM-amino acid contact are not considered when determining accessibility. Instead, PTMs may be considered, in particular, when their contribution to antigenicity is known or expected.

[0031] In preferred embodiments, the antigen, the antibody and / or the fragment comprises post-translational modifications, preferably selected from N-, O- and S-glycosylation, C-mannosylation, glycation, acylation, alkylation, butyrylation, malonylation, propionylation, sulfation, phosphorylation, ubiquitination, and smoylation.

[0032] It is understood that the PTM is included in the structure or ensemble of structures, more specifically, in silico representation thereof according to the present invention. The above non-exclusive list of PTMs refers to common and well-known PTMs.

[0033] In a more preferred embodiment, the ensemble represents the structural diversity of the antigen, the antibody, and / or the fragment at temperatures that occur in mammalian, preferably human, tissues, and / or at temperatures of 273–323K, such as 300–310K.

[0034] This embodiment provides a simulation of typical conditions in an environment where interactions related to immunity, antibody enhancement, and immune evasion occur. In a further preferred embodiment, the structures or ensembles of (a) and / or (b) are obtained from a database of three-dimensional structures by crystallography, NMR spectroscopy, electron microscopy, mass spectrometry, atomic force microscopy, and / or in silico structure prediction.

[0035] This is a non-exhaustive list of general sources of three-dimensional structural information. Anything contained in the listed databases can be obtained by any of the methodologies in the list. For example, entries in a three-dimensional structural database may be obtained, for example, by X-ray crystallography or NMR spectroscopy, or, in the case of carrying out the present invention, X-ray crystallography or NMR spectroscopy may be employed in the first step to obtain an in silico representation of a structure or ensemble of structures, which is then supplied to the first embodiment of the method as defined above.

[0036] In silico structure prediction methods are known in the art and include predictions from protein amino acid sequences and / or fragmentary or low-resolution structural information using methods such as homology modeling, integrated modeling, or artificial intelligence-based techniques like AlphaFold. See, for example, Tunyasuvunakool et al., Nature 596:590-596 (2021).

[0037] Databases of three-dimensional structures are known in this field, and examples include the Protein Data Bank (PDB), see Berman et al., Nucleic Acids Research 28:235-242 (2000), and the servers operated by the Research Collaboratory for Structural Bioinformatics (RCSB).

[0038] In a further preferred embodiment, structural sampling of the structures and / or structural exploration of the ensemble is performed by computer simulations, such as molecular dynamics simulations and / or Monte Carlo simulations, preferably on a microsecond timescale or a representative timescale of microseconds. Outlines of structural sampling methods can be found in Liwo and Scheraga, Current Opinion in Structural Biology 18:134-139 (2008), and Jorgensen and Tirado-Rives, J. Phys. Chem. 100(34):14508-14513 (1996). Monte Carlo simulations can be found in Linke M, Quoika P, Bramas B, Kofinger J, and Hummer G. Complexes: Efficient and versatile coarse-grained simulations of protein complexes and their dense solutions. ChemRxiv. Cambridge: Cambridge Open Engage; 2022. A preferred implementation of molecular dynamics simulation is described in Abraham et al., SoftwareX, 1-2:19-25 (2015).

[0039] As mentioned above, the ensemble itself is thought to provide information about the available conformations and flexibilities for a given molecule. However, further conformational sampling, whether it starts with a single structure or an ensemble, can be enhanced in silico by simulation techniques. Particularly preferred in this invention are molecular dynamics simulations. With respect to time scales, the time scale covered by the simulation corresponding to the time scale of known or expected motion available for a given molecule is preferred. As a rule of thumb, the larger the molecule, the slower the time scale of motion becomes available and the more likely it is to be performed. Consequently, the computational cost increases dramatically as the size of the system being simulated increases. This is because, firstly, the size itself increases, and secondly, the motion performed by such a system becomes slower. In the examples, we demonstrate MD simulations of an antigen-antibody system sampled on a microsecond time scale.

[0040] The preferred durations for MD simulations are 0.1–1000 μs, 0.5–100 μs, or 1–10 μs. The series of time frames obtained by MD are then used as input to the method of the first embodiment. The term "time frame" refers to the set of three-dimensional coordinates of the structure at a given point in time during the MD simulation.

[0041] It is preferable to perform MD simulations for the antigen. As further disclosed below, the antibody or its fragment may have a rigid structure or an ensemble of rigid structures (the ensemble similarly expresses the flexibility of the antibody or fragment).

[0042] On the other hand, in order to enrich the information regarding the conformational space available to the antibody or fragment, it is also conceivable to perform structural sampling of the antibody or fragment, for example, by MD simulation.

[0043] In a further preferred embodiment, the determination in step (c) is made by in silico docking the structure or member of the ensemble in (a) with the structure or member of the ensemble in (b).

[0044] As used herein, “docking” generally refers to a computational method that involves bringing two or more molecules (i.e., their in silico representations) into close proximity or enabling such molecules to approach each other; see, for example, Amaro, Biophysical Journal 114:2271-2278 (2018); Fan et al., Quant. Biol. 7:83-89 (2019); Pinzi and Rastelli, Int. J. Mol. Sci. 20:4331 (2019); and Krawczyk et al., Bioinformatics 30:2288-2294 (2014). Thus, docking simulates a binding process such as the binding of an antigen to an antibody or a fragment thereof. Whether and where such binding occurs depends on the characteristics of the said structure. Since binding suggests close proximity, docking is a means of determining accessibility according to the present invention. If a region of the antigen cannot access any region of the antibody or fragment, docking between the region and the antibody or fragment is impossible, and the region is not accessible. Preferred docking methods are described in Sikora et al., PLoS Comput Biol 17(4):e1008790 (2021), and Kim and Hummer, J.Mol.Biol.375:1416-1433 (2008).

[0045] When performing docking, it is preferable to sample all possible orientations and approaches of the molecules involved. It is also possible to focus on specific regions, such as CDR on one hand and a specific epitope on the other.

[0046] In a more preferred embodiment, the docking is rigid docking. Rigid body docking enables rapid evaluation of accessibility. While the influence of the jointed object's flexibility might seem limited, this is not actually the case. Docking, especially when using ensembles, employs multiple combinations, sampling joint options and accessibility that are not available for all conformations but only for some. This, in turn, takes flexibility into account.

[0047] It is understood that docking is performed for the aforementioned multiple combinations, preferably all combinations. For example, for an ensemble of n antigen structures and an ensemble of m antibody structures, up to n × m docking operations are performed.

[0048] In a preferred embodiment, the rigid docking is preferably performed by Monte Carlo simulation using the van der Waals radii of amino acid residues or atoms of the structure and / or ensemble. For a definition of the van der Waals radius of a residue, see below for further information.

[0049] Monte Carlo simulation, as explained above, is a method for sampling the relative orientation of two or more molecules. To define the space occupied by a given molecule and determine its accessibility, it is necessary to define the size of the constituent atoms and building blocks (such as amino acids). For this purpose, the van der Waals (vdW) radius is preferably used.

[0050] With regard to accessibility, its definition is preferably similarly based on the vdW radius. Here, it is preferable to define contact (or equivalent accessibility) as occurring when two constituent blocks, two amino acids, or two atoms are at or closer to each other than a predetermined threshold, the threshold being preferably twice the sum of their respective vdW radii. Such definitions of contact (or accessibility) thresholds can, of course, be fine-tuned. For example, the threshold could be between the sum of the vdW radii and five times the sum, for example, between 1.5 and 3 times the sum.

[0051] The term "building blocks" encompasses amino acids, PTMs, atoms, and so on. In other words, for each constituent group ("building blocks") of an antigen, antibody, or fragment thereof, accessibility can be determined at the atomic level or at the level of larger parts such as amino acids or PTMs. For example, if a PTM is a glycan, it may be represented at the atomic level of detail or at the monosaccharide level with an effective van der Waals radius corresponding to each monosaccharide. If an amino acid is modified at a smaller site, the entire modified amino acid can be represented as a single entity. For means and methods of representing residues as a whole rather than at the atomic level, see below.

[0052] The term "contact" refers to intermolecular contact, i.e., contact between an antigen and an antibody (or its fragment). Contact is a binary variable. For each building block-building block combination, or for each combination of an antigen structure and an antibody or fragment structure, contact is zero if the building block-building block distance is greater than a threshold, and contact is 1 if the distance is equal to or less than the threshold.

[0053] In a particularly preferred embodiment, accessibility is quantified by counting the contacts between the antigen and the antibody or its fragments that occur during the Monte Carlo simulation.

[0054] Preferably, the contact is (i) an atom-atom contact; and / or (ii) a residue-residue contact, preferably an amino acid residue-amino acid residue contact. As described above, in either case, it is preferable to use the vdW radius to define the contact. The vdW radius of an atom is well established and its value is available from the literature; see, for example, Batsanov Inorganic Materials 37:871-885 (2001). For the vdW radius of an amino acid residue, there is a preferred definition in Kim and Hummer, J Mol Biol 375:1416-1433 (2008). This provides a coarse-grained approach and allows for better computational efficiency. The effective vdW radius of a portion containing one or more atoms can be calculated from the vdW volume of the constituent atoms, assuming the sphericity of the residue. Further information on this point can be found in Creighton TE. Proteins: Structures and Molecular Properties. 2nd Ed. WH. Freeman and Company; New York: 1993.

[0055] Preferably, contacts involving post-translational modifications are (i) not counted, or (ii) counted only for a predefined subset of said post-translational modifications. A preferred predefined subset of PTMs is one that is present with high consistency in the presented antigen and / or shows little or no variation in its chemical identity. In this context, "consistency" may refer to conservation across species or across tissues (note that PTMs may differ from tissue to tissue even within a single species). Alternatively, there may be structural criteria for consistency, such as a clearly defined density in crystal structure or cryo-electron microscopy images.

[0056] In other words, in one preferred embodiment, only contact between amino acids or their constituent atoms is considered, and amino acid / PTM and PTM / PTM contacts are not considered. In a more preferred embodiment, accessibility in terms of step (c) is the accessibility of amino acid residues of the antigen to amino acids of the complementarity-determining region (CDR) of the antibody or fragment.

[0057] This allows us to focus on antibody or fragment regions that are thought to be particularly relevant to contact in predicting site-directed mutagenesis or immune evasion. In a more preferred embodiment, the amino acid residues of the antigen are known or predicted epitopes or parts thereof.

[0058] This allows for the consideration, to the extent possible, of knowledge regarding epitopes, preferably knowledge regarding the homologous epitopes of the antibody or fragment under consideration. In a more preferred embodiment, the antigen is a pathogen, preferably a mammalian pathogen, more preferably a human pathogen, or a disease-related antigen such as a cancer-associated antigen.

[0059] In particular, pathogens are known and expected to evolve and mutate rapidly under the selective pressure exerted by components of the immune system, such as antibodies. In a more preferred embodiment, the pathogen is a virus or a bacterium, and / or the antigen is an antigen presented on the surface of the pathogen.

[0060] In a more preferred embodiment, the virus is a coronavirus such as SARS-CoV-2, and is preferably a variant of SARS-CoV-2 of concern (VoC), such as the omicron variants including BA.1, BA.2, BA.4, and BA.5.

[0061] While the present invention is not limited in this respect, it is noteworthy that it achieves outstanding performance against SARS-CoV-2, which has attracted attention since the outbreak of the global Covid-19 pandemic.

[0062] In this regard, it is particularly preferable that the antigen is the spike protein of SARS-CoV-2. The inventors were able to demonstrate that mutations in the spike protein of SARS-CoV-2 variants, including the omicron variant, are concentrated in regions where antibody access is not hindered by glycans. More specifically, the method of the present invention was able to identify candidate mutations that, when added to omicron BA.1, further increase immune evasion. Of these, the region around residue 252 of the N-terminal domain (NTD) stands out, as this region is already deleted in the lambda variant.

[0063] In a more preferred embodiment, the mutagenic activity is associated with immune evasion, particularly response to antibodies or fragments thereof as defined above. In the embodiment, the three-dimensional structure of the Fab fragment of antibody CR3022 (pdb accession number 6W41) is used.

[0064] In a second embodiment, the present invention provides the use of a computer-generated ensemble of three-dimensional structures of antigens for predicting the site-directed mutagenic activity of an antigen. A preferred embodiment of the method of the first embodiment, where applicable, specifies a preferred embodiment of the use of the second embodiment.

[0065] Furthermore, it is preferable to use at least one in silico structure of an antibody or its fragment, preferably an ensemble thereof. A preferred method for generating ensembles is MD simulation.

[0066] In a third embodiment, the present invention provides a data processing device that includes means for carrying out the method of the first embodiment or the use of the second embodiment. Such data processing devices can be implemented as computers, personal computers, tablets, workstations, or smartphones.

[0067] In a fourth embodiment, the present invention provides a computer program including instructions which, when executed by a device of the third embodiment, cause the device to perform the method of the first embodiment or the use of the second embodiment.

[0068] In a fifth embodiment, the present invention provides a computer-readable medium which, when implemented by an apparatus of the third embodiment, includes instructions causing the apparatus to implement the method of the first embodiment or the use of the second embodiment. [Brief explanation of the drawing]

[0069] [Figure 1] SARS-CoV-2 spike mutations and accessibility scores. Mutant residues in spike variants are indicated by gray squares and symbols on the plot. The consensus epitope score from Sikora et al., op. cit., is shown by a medium light gray line and a medium light shading below it, while the AAS considering glycans is shown by a dark gray line and a dark shading below it. Light gray shading in plots up to score=1 indicates surface residues listed in Table A of the S1 text. Variants are distinguished by shades and symbols as shown in the legend. NTDs and RBDs are shown at the bottom, as are the S1 / S2 furin cleavage sites (vertical dashed lines) and S2' sites (vertical solid lines). [Figure 2] Mutations in the spike protein of SARS-CoV-2 variants are concentrated in regions with high epitope and accessibility scores. Darker regions indicate the locations of mutations with omicron BA.1 and BA.5, while lighter regions indicate the locations of mutations in previous variants. The intensity of the contrast on surfaces (A) and (B) indicates high consensus epitope scores and high AAS, respectively. [Figure 3]The accessibility scores for epitopes and antibodies distinguish between spike mutation sites and common surface residues. (A) The epitope consensus score and (B) the CDF of the AAS considering glycans are calculated for all spike residues (continuous), surface residues (dotted line), the mutation site of Omicron BA.1 (dashed dotted line), Omicron BA.5 (sparsely dashed line), and earlier variants (dashed line). The Kolmogorov-Smirnov (KS) statistics described in this document correspond to the maximum vertical gap between the CDF for the mutation site of Omicron BA.1 and surface residues, respectively. The P values ​​corresponding to the epitope score (A) and AAS (B) are 0.016 and 1.5 × 10⁻⁵, respectively. [Figure 4] Immuno-evading mutations have high accessibility scores. The CDFs of AAS are shown for mutations that have been experimentally shown to reduce binding to neutralizing antibodies (Cao et al., Nature 602(7898):657-663) (dotted line). For reference, the AAS distribution of spike surface residues (dotted line) and all residues (continuous), as well as all RBD surface residues (dashed line), are shown. The score distribution was calculated with (A) no glycans and (B) glycans. The KS statistic is the maximum vertical gap between each CDF, indicated by the arrow. [Figure 5]Mutations in new variants reduce the antibody access score on the surface of SARS-CoV-2 spikes compared to wild-type (WT) and previous variants. Spikes are plotted with brightness levels indicating AAS (white: low AAS ~ gray: high AAS). (A, E): (A) "Naive" antibody access by antibodies produced after the first WT vaccination (no prior infection) or (B) after the first infection with either variant (no prior vaccination). Here, assuming that all variants confer approximately the same "native" accessibility to antibodies, renderings (A) and (E) are plotted identically. (BD) Calculated decrease in AAS for antibodies produced in response to WT vaccination. (F) Calculated decrease in AAS for antibodies produced in response to BA.1 infection. Circles indicate a significant decrease in AAS. Regions are labeled according to the N:N-terminal epitope candidates in Sikora et al., op. cit. In these plots, the AAS of all amino acids mutated from the reference sequence, specifically those within 1 nm, was set to zero. [Modes for carrying out the invention]

[0070] Examples of the present invention are described below. Example 1 material and method Antibody accessibility score To investigate the steric accessibility of the spike surface for antibody binding, extensive rigid docking of Fab fragments of antibody CR3022 (PDB ID: 6W41) was performed.

[0071] In other words, rigid docking of the coarse-grained model was performed using the simulation procedure described herein. In the Monte Carlo docking simulation, contacts between spike residues and the complementarity determination region of Fab were counted. Structures that collided with glycans were excluded from the "glycan present" (+gly) score calculation. The counts were smoothed with a 3D Gaussian filter (σ=0.5 nm) based on the simulation start structure. The obtained contact counts were then linearly mapped onto the interval [0,1] to obtain the AAS. The 5th and 95th percentiles were used as the lower and upper limits of the linear mapping.

[0072] Light accessibility score The ray accessibility score is based on the near-isotropic illumination of the spike protein by rays emitted from around the protein, as detailed in

[14] . Glycans are considered (+gly) or not considered (-gly) during ray irradiation. Raw ray hits were used to define common surface residues (see below). For all other results described here, the ray scores were smoothed using a 3D Gaussian filter (σ=0.5 nm) and linear mapping on the interval [0,1], as in the case of AAS.

[0073] Spike surface residues Based on ray accessibility data for non-glycosylated spikes from the aforementioned book, Sikora et al. defined the surface residues of the spikes. To represent both open and closed RBDs in a single ensemble, a reference spike structure with one open RBD and two closed RBDs was used. For surface residues, a cutoff was set for a total of more than 1000 ray hits in a 2.5 μs simulation. For reference, the maximum number of ray hits was approximately 5000. By imposing a minimum threshold of 1000 hits, cracks on the spike surface were excluded from the surface definition. Therefore, the definition of surface residues used here is conservative. Except for analyses that distinguished between chains B and C with closed RBDs and chain A with an open RBD, the maximum number of ray hits across the three chains was used to define the surface accessibility of a particular residue. The analysis was extended up to residue 1206, excluding transmembrane domains.

[0074] Variant avoidance feature

[0075]

number

[0076] Variant and mutation site In our analysis, we included Omicron and the previous variants Alpha, Beta, Gamma, Delta, Epsilon, Zeta, Eta, Iota, Kappa, Lambda, Mu, B.1.620, B.1.616, and P.3. As the most recent VoC, Omicron was analyzed separately from the previous variants including Delta. The main Omicron lineage (BA.1) and the recently emerged BA.5 sub-lineage were used in the analysis. Variant sites present in multiple variants were entered only once into the combined set of "previous variants". Deletions were entered at the position of each deleted amino acid. Insertions were counted only once at the position of the insertion. In the analysis, only surface residues were included because mutations within the spike are unlikely to be caused by avoidance of the antibody immune response. The same definition of the surface as used for the calculation of AAS was used.

[0077] Statistical evaluation To evaluate the significance of the difference in the distribution of epitope scores and accessibility scores, a two-sample KS test was used. For this, the KS statistic KS = maxS|P(s < S) - Q(s < S)| was determined as the extreme distance of the CDFs of the scores P(s < S) of the surface sites and the scores Q(s < S) of each set of variant sites, where the sample sizes are the number of surface residues and the number of variant sites in each set, respectively. The KS statistic, sample size, and corresponding P-value in the two-sample KS test are reported.

[0078] Example 2 Results Spike mutations concentrate in regions of predicted epitope candidates In the aforementioned book, Sikora et al. used MD simulations of membrane-immobilized SARS-CoV-2 spikes and bioinformatics tools to identify candidate epitopes on the spikes. We defined a consensus epitope score that integrates information on the surface accessibility of glycosylated spikes, their structural stiffness, sequence mutations at the time of pandemic outbreaks, and sequence-based bioinformatics predictors of epitopes. Applying this to the MD simulation trajectories of the spikes, we identified nine distinct regions on the spike surface with high epitope scores. Given the steady emergence of new SARS-CoV-2 variants, we were able to validate these computational predictions by comparing the reported mutation sites in the variant spike proteins with the predicted epitopes. Here, we assumed that spike surface mutations are largely caused by antibody-mediated immune evasion. Throughout this study, omicron variants and earlier variants were treated separately, as defined in the methodology. By comparing the mutation sites of SARS-CoV-2 variants with regions that had high predicted epitope scores, we found that the first eight of the nine predicted epitope candidates were influenced by at least one mutation in Omicron BA.1 or earlier variants (Figures 1 and 2A, C). While many mutations were concentrated in RBD and NTD regions, significant mutational activity was also observed outside these regions. To quantitatively evaluate our epitope predictions in light of variant mutational activity, we compared the consensus epitope scores of mutated residues on the spike surface with those of all surface residues. The cumulative distribution function (CDF) of the consensus epitope scores showed a significant shift towards higher scores for mutated residues in the spike variant compared to all spike residues (Figure 3A). That is, mutations occurred with high probability in regions identified as epitopes with a high probability of antibody binding. The score distribution of mutation sites also shifted to significantly larger values ​​compared to the surface residues of the spike (see Example 1 for the definition of surface residues).The P-values ​​of the KS statistics for omicron BA.1 and earlier variants regarding surface residues were 0.016 and 0.010, respectively, and are therefore significant. A high epitope consensus score (Sikora et al., op. cit.) correlates with the occurrence of mutations in the spike protein of SARS-CoV-2 variants.

[0079] Antibody accessibility score as a predictor of mutation occurrence in related variants The inventors aimed to construct a simpler and more powerful predictor of variant activity in variant spikes based solely on surface accessibility, i.e., without including structural rigidity, sequence conservation, or sequence signature. We compared three surface accessibility scores—"ray," "antibody accessibility score (AAS, referred to as "docking" in Sikora et al., op. cit.)"—with and without surface glycosylation. The ray score quantifies accessibility in terms of surface illuminance due to diffuse light. See Example 1 above and Sikora et al., op. cit. AAS measures the extent to which each residue is accessible to the antibody fragment probe. SASA defines the solvent-accessible surface by a spherical probe (radius: 0.14 nm).

[0080] All surface accessibility scores were determined for the complete glycosylated spike structure collected during 4 × 2.5 μs MD simulations. This accounts for dynamic shielding by glycosylation. Below, we evaluate the different scores of variant mutation sites on the spike surface and compare them to the scores of typical surface residues. Here, we assume that mutations within the spike are caused by factors other than antibody binding. For the definition of typical spike surface residues, we used ray scores that do not include surface glycosylation. Compared to conventional surface definitions based on SASA, our ray analysis does not consider deep cracks as surface exposures. Therefore, the estimates of the difference in score distribution between variant residues and surface residues are quite conservative in the sense that the difference in score distribution could be even larger with a more generous definition. Consistently across variants, AAS surprisingly appeared to be the best score (Figure 3B, Table 1 below).

[0081] [Table 1] The table above shows the KS statistics and corresponding P-values ​​obtained for the distribution of ray scores, antibody accessibility scores (AAS), and SASA scores when comparing common surface residues (S; Table A in the S1 text) with surface mutation sites of Omicron BA.1 (Table B in the S1 text), Omicron BA.5, and earlier variants (EV). Scores were calculated with and without glycans. The sample size N consisted of 29 mutations in Omicron BA.1, 30 mutations in Omicron BA.5, 51 mutations in earlier variants, 19 mutations different between BA.1 and BA.5, and 620 common surface residues.

[0082] We used the KS statistic to quantify the power of different scores to distinguish between variant sites on the spike surface and common surface residues of the spike. The p-value of the KS test for AAS using glycans for all surface residues was considerably better than the previous epitope score. The p-value for omicron BA.1, analyzed using AAS and glycans, was 1.5 × 10⁻⁶. -5and the P-value of the previous variant was 2.3×10 -8 In contrast, the simpler ray score had slightly lower discriminative power (Table 1). The commonly used SASA score was not the most suitable for distinguishing mutated surface residues from common surface residues, and the P-value for Omicron BA.1 was as low as 0.31.

[0083] Omicron BA.5 had a slightly lower AAS value distribution compared to Omicron BA.1 and previous variants (Figure 3B). However, notably, the set of residues that differ between the BA.1 and BA.5 variants shows a very large shift in the direction of higher antibody accessibility in all measurement criteria including glycan shielding (BA.5 - BA.1 - S in Table 1).

[0084] Lack of glycan coating is a determinant of mutational activity Next, we wondered whether glycosylation actually plays a central role in antibody-based immune steering, as assumed in the construction of the original epitope score by Sikora et al., supra. The CDFs of AAS and ray scores were calculated without including glycosylation. In Table 1, the KS statistic and corresponding P-values are compared as measures of the discriminative power between the mutated sites and common surface sites on the spike surface in the major variants. Considering glycosylation, it was found that the discriminative power of AAS was significantly improved and that of the ray score was somewhat improved. For the mutated sites of Omicron BA.1 and previous variants, when evaluated by AAS for all surface sites, including glycosylation, the P-value improved from 0.02 to 1.5×10 -5 to 2.1×10 -6 from 2.3×10 -8 For the ray score including glycosylation, the P-value for the mutated sites of Omicron BA.1 improved from 8.0×10 -4 to 3.1×10 -5 and for the previous variant from 9.0×10 -7 to 3.8×10 -7This resulted in an improvement. By considering dynamic glycan shielding in this way, the discriminative power of AAS was significantly improved, and the discriminative power of ray scores was also slightly improved.

[0085] Effects of protein and glycan dynamics Next, we wondered whether sampling the dynamics of proteins and glycans is essential for identifying variant mutation sites from common surface residues. For this purpose, we calculated ray scores for static initiation structures in MD simulations, both with and without glycation. Here, we added glycans to the protein structures in a post-processing stage using the software GlycoSHIELD (doi.org / 10.1101 / 2021.08.04.455134). GlycoSHIELD is used to add glycan structures to proteins in a predefined sequence by drawing glycan conformation isomers from a simulation library. Interestingly, the single non-glycosylated spike protein structure analyzed by light scoring was sufficient to distinguish between mutant residues and surface residues in Omicron BA.1 and earlier variants, but not in Omicron BA.5. The P-value for Omicron BA.1 was 0.021, for Omicron BA.5 it was 0.47, and for earlier variants it was 8.3 × 10⁻⁶. -5 This score was significant for Omicron BA.5 and could be further improved by sampling glycan motion in Omicron BA.1 and earlier variants. Using GlycoSHIELD (cutoff 3.5 Å, coarse-grained mode, glycan type defined at doi.org / 10.1101 / 2021.08.04.455134), we constructed an ensemble of 160 spike chains with different glycan conformation isomers attached to a static protein structure. This ensemble yielded 1.9 × 10⁶ values ​​for Omicron BA.1, Omicron BA.5, and earlier variants, respectively. -4 , 0.039, and 2.9 × 10 -6The following p-values ​​were obtained. As expected, the best performance was achieved by sampling the movement of both proteins and glycans from the MD simulation structure (Table 1), with p-values ​​of 3.1 × 10⁻⁶ each. -5 , 9.6×10 -3 , 3.8×10 -7 This trend is expected to be maintained in AAS as well. While expensive MD simulations are required to obtain the best predictive power, mean-field approaches like GlycoSHIELD guarantee accurate predictions at minimal computational cost. This has proven valuable for high-throughput studies across a wide range of sequence and structural models of viral and nonviral targets.

[0086] Immune-evading mutations have a high accessibility score. Recent experimental studies (Cao et al., Nature 602(7898):657-663(2022)) identified a series of spike mutations that significantly weaken neutralizing antibody binding. In Figure 4, these mutation sites show a considerably higher AAS than typical surface residues. Considering glycosylation, this shift is 4.6 × 10⁻⁶. -5 This results in a p-value. Even when the analysis is limited to RBDs, the mutation sites show a significant shift to higher AAS values ​​compared to RBD surface residues (shown as dotted and dashed lines in Figure 4, respectively). We conclude that the relatively simple surface accessibility score presented here can reliably identify spike mutation sites important for immune evasion.

[0087] Omicron differs from previous variants in cumulative score. We also wondered whether Omicron differed from previous variants in terms of spike surface mutations. The Omicron variant is phylogenetically distant from other major variants and does not belong to the previous VoC antigen cluster. However, the overall distribution of AAS for Omicron BA.1 and the previous variants is similar (Figure 3B). In contrast, the AAS distribution for BA.5 is slightly shifted to lower values ​​for most scores. Looking more closely, a similar accessibility score distribution is observed for all variants. Therefore, the accessibility of individual mutant residues in Omicron is, on average, not significantly different from the accessibility in the previous variants. What is different, however, is the larger total number of mutations in the two Omicron variants considered here, which is expected to have an overall impact on immune evasion.

[0088] Possibility of avoiding SARS-CoV-2 variants To quantify the overall effect of mutations in variants, we introduced a “avoidability” score. The avoidability score F aims to quantify the extent to which mutations in variants cause loss of binding to antibodies enhanced against WT spikes, similar to current vaccines. Motivated by our observation that mutation sites are concentrated in regions with high AAS, F measures the weighted proportion of the antibody-accessible surface modified by the mutation with respect to a reference sequence used, for example, in vaccination. Considering the fact that mutations disrupt the structure and chemical properties around the mutation site, we extend the effect of the mutation to a radius of 1 nm centered on the mutation site. Thus, the avoidability score F is defined in Equation 1 (Example 1) as the sum of AAS S(i) across all residues i within 1 nm of the mutation site. For brevity, we have not attempted to limit mutations by changes in amino acid chemistry, which are clearly important for specific binding. For example, extensive screening of antibodies has revealed that their ability to neutralize omicron variants is subtly dependent on mutations in mapped epitopes. According to the definition, a WT spike score is 0, and the highest avoidability score is 1.0. Based on this definition, Omicron BA.1 and BA.5 were found to have substantially higher avoidability scores F compared to other past and present VoC alpha, beta, gamma, and delta (Table 2).

[0089] [Table 2] Furthermore, the Omicron variant differs from previous variants by having a significantly higher avoidance score for chain A, where the RBD is open in our model. Note that the RBD opening of chain A disrupts the symmetry between chains B and C, which is reflected in the slightly different avoidance scores of the individual chains in Table 2. The improved avoidance of Omicron BA.5 against the WT antibody is noteworthy, given that the individual distributions are slightly shifted to lower scores (Figure 3B). The large value of F=0.52 for chain A of Omicron BA.1, calculated using AAS and glycans, is greater than the F≒0.42 for chains B and C, due to a relatively larger number of mutations localized in the RBD region exposed when the RBD opens for binding to the ACE2 receptor. In contrast, the avoidance scores of the three chains are fairly even in the alpha, beta, gamma, and delta variants. Given that numerous mutations in the omicron spike did not significantly improve its binding affinity to ACE2, our premise that antibody-based immunodefence promotes the selection of spike mutations appears to hold true even for functionally important RBD mutations. We investigated whether the recent omicron variant BA.5 is predicted to be more favorable to immune evasion than BA.1 after WT vaccination. Table 2 shows that BA.5 does not increase evasion from WT vaccination. Next, we examined whether antibody evasion from BA.1 infection, in addition to WT vaccination, could be a driving force for the newly emerging variants BA.2 and BA.5. We evaluated the evasion of BA.5 against the BA.1 variant and found that BA.5 exhibited considerably higher evasion with BA.1 as the baseline. Therefore, our data suggest that, in addition to factors such as infectivity and incubation period, chemical differences in accessible residues compared to previously dominant variants support antibody evasion. To illustrate this point, the spike protein was plotted in grayscale according to AAS (Figure 5). In Figure 5, the AAS was set to zero for all residues within 1 nm of an amino acid different from the reference sequence.Panel AD in Figure 5 shows the reduction in AAS due to mutations in antibodies produced against WT vaccination (Figure 5A), and panel EF shows the reduction in AAS due to mutations in antibodies produced against BA.1 infection (Figure 5E). (Note that we assume all infections or vaccinations in “naive” patients who have never been exposed to the SARS-CoV-2 spike are within the wild-type (native) range. Therefore, panels 5A and 5E are identical.) Spike variants BA.1 and BA.5 show stronger suppression of AAS relative to WT than delta (Figure 5B-D). Furthermore, BA.5 causes substantial suppression of AAS relative to BA.1 (Figure 5F).

Claims

1. A computer implementation method for predicting site-directed mutagenicity of an antigen, wherein the method is: (a) Providing at least one three-dimensional structure of the antigen, preferably an ensemble of the three-dimensional structures of the antigen, in silico; (b) Providing in silico at least one three-dimensional structure of at least one antibody or fragment thereof that is binding to or specific to the antigen, preferably an ensemble of three-dimensional structures of the antibody or fragment thereof, (c) The step of determining in silico the accessibility of the structure of (a) to the structure of (b), wherein higher accessibility indicates higher mutagenic activity, particularly under the selective pressure exerted by the antibody or its fragment, and the determination is made by in silico docking the structure of (a) or a member of the ensemble with the structure of (b) or a member of the ensemble. This provides a method for predicting the site-directed mutagenesis activity.

2. The method according to claim 1, wherein the antigen, the antibody and / or the fragment thereof comprises post-translational modifications, preferably selected from N-, O- and S-glycosylation, C-mannosylation, glycation, acylation, alkylation, butyrylation, malonylation, propionylation, sulfation, phosphorylation, ubiquitination, and smoylation.

3. The method according to claim 1 or 2, wherein the ensemble expresses the structural diversity of the antigen, the antibody, and / or the fragment thereof at temperatures occurring in mammalian, preferably human, tissue, and / or at temperatures of 273 to 323 K, such as 300 to 310 K.

4. The method according to any one of claims 1 to 3, wherein the structure or ensemble of (a) and / or (b) is obtained from a database of three-dimensional structures by crystallography, NMR spectroscopy, electron microscopy, mass spectrometry, atomic force microscopy, and / or in silico structure prediction.

5. The method according to claim 4, wherein stereostructural sampling of the structure is performed and / or the stereostructural sampling of the ensemble is extended by computer simulations such as molecular dynamics simulations and / or Monte Carlo simulations, preferably on a timescale of microseconds or a representative timescale of microseconds.

6. The method according to any one of claims 1 to 5, wherein the docking is rigid docking.

7. The method according to claim 6, wherein the rigid docking is performed by Monte Carlo simulation, preferably using the van der Waals radii of amino acid residues or atoms of the structure and / or ensemble.

8. The method according to claim 7, wherein accessibility is quantified by counting the contacts between the antigen and the antibody or its fragments that occur during the Monte Carlo simulation.

9. The aforementioned contact, (i) Atomic-atomic contact; and / or (ii) Residue-residue contact, preferably amino acid residue-amino acid residue contact The method according to claim 8.

10. Contacts involving post-translational modifications are (i) Not counted; or (ii) Only a predefined subset of the post-translational modifications is counted, The method according to claim 8 or 9.

11. The method according to any one of claims 1 to 10, wherein the accessibility with respect to step (c) is the accessibility of the amino acid residues of the antigen to the amino acids of the complementarity-determining region (CDR) of the antibody or fragment thereof.

12. The method according to any one of claims 1 to 11, wherein the antigen is a pathogen, preferably a mammalian pathogen antigen, more preferably a human pathogen antigen, or a disease-related antigen such as a cancer-associated antigen.

13. The use of a computer-generated ensemble of three-dimensional structures of an antigen to predict its site-directed mutagenic activity.

14. A computer program that, when executed by a data processing device, includes an instruction causing the device to perform the method described in any one of claims 1 to 12.

Images