Method for predicting immunogenicity of epitope against HLA allele based on sequence similarity and device using same
By clustering HLA alleles based on sequence similarity and determining weights, the method effectively predicts immunogenicity for HLA alleles not in training data, enhancing prediction accuracy and reducing experimental requirements.
Patent Information
- Application Number
- KR1020220148081
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2022-11-08
- Publication Date
- 2026-11-11
- Estimated Expiration
- 2042-11-08
AI Technical Summary
Existing methods and programs for predicting the immunogenicity of epitopes and neoepitopes are inefficient and inaccurate when dealing with HLA alleles not included in training data, often requiring costly and time-consuming experiments, and fail to consider additional characteristics like epitope-HLA binding stability and phosphorylated epitope-HLA binding.
A method that clusters HLA alleles based on sequence similarity, determines weights of similar alleles, and predicts immunogenicity by summing the confirmed immunogenicity of these alleles, using a sequence clustering algorithm and considering factors like race and population to form clusters.
Enables accurate prediction of immunogenicity for HLA alleles not in training data, improving prediction accuracy and reducing the need for additional experiments.
Smart Images

Figure 112022118767211-PAT00004_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a method for predicting the immunogenicity of an epitope for a specific HLA allele based on sequence similarity and an apparatus using the same. More specifically, the invention relates to a method for calculating a predicted value of the immunogenicity of an epitope for an HLA allele that cannot be directly predicted based on sequence similarity between an HLA allele that cannot be predicted immunogenicity for the epitope and HLA alleles that can be predicted immunogenicity, and an apparatus using the same. Background Technology
[0002] Although various methods and / or programs are utilized to predict the immunogenicity of epitopes and neoepitopes in an in silico manner, the experimental data used for training by each method and / or program is limited. When predicting the immunogenicity of HLA alleles not included in the training data, these methods and / or programs cannot predict the immunogenicity of the target epitopes for those specific HLA alleles, requiring the acquisition of data through individual experiments. However, such methods may be inefficient due to high time and cost; furthermore, excluding HLA alleles not included in the training data from immunogenicity prediction may make it impossible to consider those specific HLA types. Therefore, there is a need for methods and / or programs for predicting the immunogenicity of epitopes and neoepitopes that can predict the immunogenicity of HLA types for HLA alleles not included in the training data without additional experiments.
[0003] sNebula, one of the proposed solutions to address this, predicts immunogenicity between epitopes and HLA alleles that are not connected (i.e., not learned) in the network through a bipartite network composed of two components: an epitope and an HLA allele. It utilizes existing learned information between sequence-similar epitopes or sequence-similar HLA alleles that are connected to the corresponding epitope or HLA allele. However, sNebula has limitations in that it does not consider other characteristics associated with HLA alleles, such as epitope-HLA binding stability and phosphorylated epitope-HLA binding, in addition to epitope-HLA allele binding, when predicting immunogenicity.
[0004] Another approach, KNN (K Nearest Neighbor), vectorizes the remaining columns of matrix-form data to predict missing values based on the distance between vectors, specifically for columns where some actual values are missing. However, because KNN utilizes information from other features to predict missing values for each feature, it has a limitation in that dependencies arise between the inferred values when considering various immunogenicity characteristics individually. The problem to be solved
[0005] The present invention proposes a method for reasonably predicting the immunogenicity of epitopes for HLA alleles not included in training data. means of solving the problem
[0006] A method for predicting the immunogenicity of an epitope for an HLA allele according to the present invention may include the steps of: measuring sequence similarity between an HLA allele to be predicted for immunogenicity and other HLA alleles within a cluster including said HLA allele to be predicted for immunogenicity; determining weights of other HLA alleles within the cluster based on said measured sequence similarity; confirming the immunogenicity of the epitope for other HLA alleles within the cluster; and predicting the immunogenicity of the epitope for the HLA allele to be predicted for immunogenicity based on said confirmed immunogenicity of the epitope for the HLA allele and said determined weights of the HLA alleles.
[0007] In the method for predicting the immunogenicity of an epitope for an HLA allele according to the present invention, the HLA allele may be an HLA allele that has not been learned or obtained and therefore cannot predict the immunogenicity of the epitope through a specific program or a specific method.
[0008] In the method for predicting the immunogenicity of an epitope for an HLA allele according to the present invention, the HLA allele cluster may be a cluster formed by collecting publicly available HLA alleles and using a sequence clustering algorithm or software based on sequence similarity.
[0009] In the method for predicting the immunogenicity of an epitope for an HLA allele according to the present invention, the HLA allele cluster may be clustered by further considering at least one of race and population.
[0010] In the method for predicting the immunogenicity of an epitope for an HLA allele according to the present invention, the sequence clustering algorithm or software may be a sequence clustering algorithm or software that targets the entire length of the HLA sequence or extracts a pseudo-sequence region that forms a core binding with the epitope peptide and clusters by measuring sequence similarity only for the sequence of that region.
[0011] In the method for predicting the immunogenicity of an epitope for an HLA allele according to the present invention, the sequence clustering algorithm or software may be a sequence clustering algorithm or software that clusters by comparing the degree of agreement of the sequence itself, or clusters HLAs with similar constituent amino acid properties by utilizing physicochemical properties or a BLOSUM matrix.
[0012] In the method for predicting the immunogenicity of an epitope for an HLA allele according to the present invention, the criteria for clustering collected HLA allele sequences may be at least one of the degree of sequence similarity, the distance between vector coordinates (Euclidean distance), the vector angle similarity (cosine similarity), and a predetermined number.
[0013] In the method for predicting the immunogenicity of an epitope for an HLA allele according to the present invention, the step of predicting the immunogenicity of an HLA allele may be a step of predicting the immunogenicity of the HLA allele (A) using only some of the HLA allele clusters based on the weights of the determined HLA allele clusters.
[0014] In the method for predicting the immunogenicity of an epitope for an HLA allele according to the present invention, a cluster including an HLA allele to be predicted immunogenicity may include HLA alleles within a certain distance from the HLA allele to be predicted immunogenicity.
[0015] In the method for predicting the immunogenicity of an epitope for an HLA allele according to the present invention, a cluster containing an HLA allele to be predicted for immunogenicity may include a predetermined number of HLA alleles in order of proximity to the HLA allele to be predicted for immunogenicity.
[0016] A computer-readable recording medium for predicting the immunogenicity of an epitope for an HLA allele according to the present invention may have a computer program recorded thereon for executing at least one of the methods described above.
[0017] An apparatus for predicting the immunogenicity of an epitope for an HLA allele according to the present invention may include an input / output module that receives an HLA allele for which immunogenicity is to be predicted and outputs the predicted immunogenicity, a memory in which a disclosed HLA allele sequence, a clustering algorithm, or software is stored, and a processor that measures the sequence similarity between the HLA allele for which immunogenicity is to be predicted and other HLA alleles within a cluster including the HLA allele for which immunogenicity is to be predicted, determines weights of other HLA alleles within the cluster based on the measured sequence similarity, confirms the immunogenicity of an epitope for other HLA alleles within the cluster, and predicts the immunogenicity of an epitope for the HLA allele for which immunogenicity is to be predicted based on the immunogenicity of the epitope for the confirmed HLA alleles and the determined weights of the HLA alleles. Effects of the invention
[0018] According to the present invention, the immunogenicity of an epitope for an HLA allele not included in the training data can be reasonably predicted. Brief explanation of the drawing
[0019] FIG. 1 is a flowchart of a device for predicting the immunogenicity of an HLA allele based on sequence similarity according to one embodiment of the present invention. Figure 2a shows sequence information of HLA alleles and similar HLA alleles collected from the IMGT database, and Figure 2b shows the values converted into vectors. Figure 3a shows HLA alleles with sequences similar to the target HLA allele in two dimensions, and Figure 3b shows a table of values inferred for the immunogenicity of the target epitope of an unlearned HLA allele by confirming the immunogenicity values of the target epitope predicted for similar HLA alleles and summing them with weights based on the similarity distance to the target HLA allele. FIG. 4 is a block diagram of a device for predicting the immunogenicity of an epitope for an HLA allele based on sequence similarity according to one embodiment of the present invention. Specific details for implementing the invention
[0020] The following examples are provided solely to illustrate the invention more specifically, and it will be obvious to those skilled in the art that the scope of the invention is not limited by these examples according to the gist of the invention. It should be understood that the invention includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the technology described below.
[0021] In terms used herein, singular expressions should be understood to include plural expressions unless the context clearly indicates otherwise, and terms such as “comprising” should be understood to mean that the described features, number, step, action, component, part, or combination thereof exist, and not to exclude the existence or addition of one or more other features, number, step, action, component, part, or combination thereof. The term “and / or” includes a combination of multiple related described items or any of the multiple related described items.
[0022] In performing the method or operation method, each process / step constituting the method may occur differently from the specified order unless a specific order is clearly indicated in the context. That is, each process / step may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.
[0024] Various factors contributing to immunogenicity can be utilized to predict immunogenic epitopes and neoepitopes. Since each factor may act at different stages of the immune system mechanism, various experiments are required to measure the extent of action of relevant factors at each stage in order to assess the impact of each factor on immunogenicity. However, in most cases, experiments are conducted using only a subset of HLA (human leukocyte antigen) alleles due to limitations in available samples, experimental methods, time, and / or cost, resulting in limited experimental data. When using publicly available tools or training new machine learning methods to predict the influence of each factor contributing to immunogenicity on a specific epitope, it may be impossible to predict the immunogenicity of HLA alleles that are not included in the trained or acquired data.
[0025] In the following, we describe in detail a method for constructing clusters of publicly available HLA alleles based on sequence similarity, and for predicting the immunogenicity of HLA alleles that have not been trained or obtained and whose immunogenicity for epitopes cannot be predicted by specific programs and / or methods, using clusters constructed based on sequence similarity.
[0027] FIG. 1 is a flowchart of a device for predicting the immunogenicity of an HLA allele based on sequence similarity according to one embodiment of the present invention.
[0028] An apparatus for predicting the immunogenicity of an HLA allele based on sequence similarity according to one embodiment measures the sequence similarity between the HLA allele to be predicted for immunogenicity and other HLA alleles in a cluster including the HLA allele to be predicted for immunogenicity (S110). The HLA allele to be predicted for immunogenicity may be an HLA allele that has not been learned or obtained and therefore cannot predict immunogenicity for an epitope through a specific program or / and a specific method.
[0029] According to one embodiment, an HLA allele cluster (or cluster) may be formed by collecting disclosed HLA alleles and clustering the collected HLA alleles using a sequence clustering algorithm or software. The disclosed HLA alleles may be collected, for example, from the IMGT (ImMunoGeneTics) / HLA database (http: / / www.ebi.ac.uk / imgt / hla). The sequence clustering algorithm or software may cluster HLA alleles based on sequence similarity. According to one embodiment, the sequence clustering algorithm or software may cluster by measuring sequence similarity over the entire length of the HLA allele sequence, or by extracting pseudo-sequence regions that bind to epitope peptides and measuring sequence similarity only over the sequences of those regions. According to one embodiment, the sequence clustering algorithm or software may cluster by comparing the degree of agreement of the sequences themselves, or by utilizing physicochemical properties or a BLOSUM matrix to cluster HLA alleles with similar constituent amino acid properties. The criteria for clustering collected HLA allele sequences may be at least one of the following: the degree of sequence similarity (e.g., above a specific value), the distance between vector coordinates (Euclidean distance), the vector angle similarity (cosine similarity), or a fixed number (e.g., N in order of highest sequence similarity). As an example, a cluster containing an HLA allele for which immunogenicity is predicted may include all or at least some of the HLA alleles within a certain distance from the HLA allele for which immunogenicity is predicted. As another example, a cluster containing an HLA allele for which immunogenicity is predicted may include a fixed number of HLA alleles in order of proximity to the said HLA allele for which immunogenicity is predicted.
[0030] According to one embodiment, since the types and frequencies of HLA alleles possessed by each race / group differ, HLA allele clusters may be clustered by further considering at least one of race or group. For example, HLA allele clusters may be clustered based on sequence similarity by distinguishing races / groups such as Asians, Europeans, Africans, and Latin Americans, and limiting them to the types of HLA allele sequences found in the corresponding races / groups.
[0031] A device for predicting the immunogenicity of an epitope for an HLA allele based on sequence similarity according to one embodiment determines the weights of other HLA alleles within a cluster containing the HLA allele to be predicted based on the measured sequence similarity (S120). For example, the weight of an HLA allele with high sequence similarity can be determined to be high, and the weight of an HLA allele with low sequence similarity can be determined to be low.
[0032] An apparatus for predicting the immunogenicity of an epitope for an HLA allele based on sequence similarity according to one embodiment confirms the immunogenicity of an epitope for other HLA alleles within a cluster containing the HLA allele to be predicted (S130). The immunogenicity of the epitope for HLA alleles can be confirmed through a specific program or / and a specific method. The immunogenicity confirmed through the specific program or / and a specific method can be expressed as a numerical value.
[0033] An apparatus for predicting the immunogenicity of an epitope for an HLA allele based on sequence similarity according to one embodiment predicts the immunogenicity of an epitope for an HLA allele based on the immunogenicity of the epitope for the identified HLA alleles and the weights of the determined HLA alleles (S140). According to one embodiment, an apparatus for predicting the immunogenicity of an epitope for an HLA allele based on sequence similarity may use only some of the HLA alleles included in the cluster.
[0034] According to one embodiment, the immunogenicity of an epitope for an HLA allele can be determined as follows. To distinguish between the HLA allele included in the HLA allele cluster and the HLA allele for which immunogenicity is to be determined, the HLA allele for which immunogenicity is to be determined is denoted as a or HLA allele(a), and the HLA allele included in the HLA allele cluster is denoted as k or HLA allele(k). The immunogenicity between HLA allele(a) and the epitope is f a When expressed as (x), f a (x) can be expressed as [Equation 1].
[0035]
[0036] Here, f k (x) represents the immunogenicity between the HLA allele(k) included in the HLA allele cluster and the epitope, and n is all or part of the number of HLA allele(k) included in the HLA allele cluster. W k is the sequence similarity (S) between the HLA allele (k) included in the HLA allele cluster and the HLA allele (a) for which immunogenicity is to be determined. k,a ) or vector distance between sequences (D k,a It is a weight based on ).
[0037] According to one embodiment, the method for converting immunogenicity calculation values into weight values for summing based on sequence similarity may use [Equation 2] or [Equation 3]. When using a similarity scale where a higher value indicates greater similarity, the sequence similarity (S) between the HLA allele (k) included in the HLA allele cluster and the HLA allele (a) for which immunogenicity is to be obtained, as in [Equation 2], k,a ) sequence similarity (S) between the HLA allele (k) included in all HLA allele clusters and the HLA allele (a) for which immunogenicity is to be determined k,a Weights can be obtained by dividing by the sum of ). If a similarity measure is used where a lower value indicates greater similarity, weights can be obtained by dividing the reciprocal of the vector distance between the sequences of the HLA allele (k) included in the HLA allele cluster and the HLA allele (a) for immunogenicity, as shown in [Equation 3], by the sum of the reciprocals of the vector distances between the sequences of the HLA allele (k) included in all HLA allele clusters and the HLA allele (a) for immunogenicity.
[0038]
[0039]
[0040] Depending on the results of the inference performance evaluation, a weight calculation method can be set so that differences due to sequence similarity are reflected more or less in the weights. Here, the inference performance evaluation can be performed by assuming that the experimental value of the immunogenicity of the epitope for HLA alleles, which has been measured or can be predicted using a prediction tool, is the correct answer, and comparing how closely an indirectly predicted value approximates the correct answer by weighting the predicted values of immunogenicity between the epitope and other HLA alleles within a cluster that have sequences similar to the HLA allele corresponding to the correct answer.
[0041] According to one embodiment, if the range of immunogenicity prediction values calculated using a program / calculation method includes negative and positive numbers, or has a large range ranging from a decimal less than 1 to an exponent greater than 10, then even if an HLA allele has low sequence similarity, if the immunogenicity value is large, it may be significantly reflected in the final sum by a weighted sum method. Alternatively, a method may be applied in which immunogenicity values are calculated in advance for all epitopes and all HLA alleles that can be calculated in the program and converted into percentiles, weighted sums are performed based on the converted percentile values, and finally the summed percentile is converted back into a prediction value.
[0043] In the following, we confirm whether the method for predicting the immunogenicity of HLA alleles based on sequence similarity according to one embodiment of the present invention is an effective method, and demonstrate that the method according to the present invention is effective compared to conventional methods.
[0044] Specifically, verification was performed using sequence information of HLA-A alleles among HLA class I, for which the most data has been established. First, sequence information of HLA-A alleles and epitope-HLA binding information were collected from the IEDB (Immune Epitope Database). For HLA alleles for which epitope-HLA binding information is publicly available in the IEDB, immunogenicity can be predicted in most programs such as netMHCpan, netMHCstabpan, netMHCphospan, and PRIME. Since it is difficult to evaluate the predicted values according to the present invention for HLA alleles for which immunogenicity prediction values cannot be calculated through the program, some HLA alleles were assumed to be unpredictable in terms of immunogenicity, and the performance of the present invention was compared by comparing the difference between the predicted value according to the present invention and the actual calculated immunogenicity prediction value for the corresponding HLA allele using the immunogenicity prediction values of other HLA alleles. [Table 1] below shows the number of HLA alleles and epitope binding information assumed to be learned / unlearned among the collected HLA alleles.
[0045] Predictable Unpredictable total HLA allele 10 16 26 Epitope 2,614 211 2,825
[0046] A total of 2,825 epitope-HLA-A allele binding data were collected from the IEDB. This includes epitope binding data for 26 HLA-A alleles. Of these, 2,614 (92%) of the 2,825 binding data were for 10 HLA allele-epitope binding data with high frequency in the population, and 221 (8%) were for 16 HLA allele-epitope binding data with low frequency. Accordingly, assuming that the 16 HLA allele-epitope binding data with low frequency are unpredictable by the comparison program, the prediction performance of the method according to the present invention, the previously disclosed method using sNebula for prediction, and the universally used KNN method was compared based on the predicted values calculated for the corresponding epitopes for the remaining 10 HLA alleles with high frequency.
[0047] More specifically, assuming that it is impossible to predict the immunogenicity of the HLA-A*30:01 allele among the 221 epitope-HLA allele binding information used in the prediction performance evaluation through the program, the immunogenicity value of A*30:01 was predicted (or inferred) from the program prediction values of A*03:01, A*11:01, A*32:01, A*01:01, A*32:02, and A*26:01 alleles that have similar sequences.
[0048] Figure 2a shows sequence information of HLA alleles and similar HLA alleles collected from the IMGT database, and Figure 2b shows the values converted into vectors.
[0049] Specifically, Figure 2a shows the actual amino acid sequence of the HLA-A*30:01 allele collected from the IMGT database and the differences in sequence between the A*03:01, A*11:01, A*32:01, A*01:01, A*32:02, and A*26:01 alleles that have similar sequences.
[0050] Figure 2b shows an HLA allele sequence composed of 20 amino acids (letters) that has been converted into a vector by quantifying it into values representing physicochemical properties or amino acid properties (BLOSUM), and then plotted on a coordinate plane / space after dimensionality reduction through PCA (Principal Component Analysis) transformation. Methods for quantifying the sequence may include sequence identity (%identity) with respect to a genome map or population standard sequence, physicochemical properties of the HLA allele constituent amino acids (e.g., hydrophobicity score), and BLOSUM matrix. The entire HLA amino acid sequence or the amino acids at 48 pseudo-sequence positions located in the binding pockets that bind to epitopes can be quantified into a one-dimensional or two-dimensional vector. Sequence similarity between HLA alleles can be measured based on coordinate distance and vector similarity.
[0051] Figure 3a shows HLA alleles with sequences similar to the target HLA allele in two dimensions, and Figure 3b shows a table of immunogenicity values for the target epitope of the target HLA allele inferred through the weighted sum of their immunogenicity for the target epitope.
[0052] Figure 3a shows, as an example, HLA alleles learned in a two-dimensional plane in green and unlearned target HLA alleles in red. Referring to Figure 3a, the HLA alleles forming a cluster may vary depending on the position of the target HLA allele, the HLA-A*30:01 allele, in the two-dimensional plane. For example, HLA alleles within a certain distance range centered on the HLA-A*30:01 allele predicting immunogenicity can be determined as a single cluster. Alternatively, a fixed number of HLA alleles in order of shortest distance from the HLA-A*30:01 allele predicting immunogenicity can be determined as a single cluster. Regardless of the number of HLA alleles included in the cluster, through weighted summation based on sequence similarity, the predicted values of HLA alleles with more similar sequences can be reflected more significantly, while the predicted values of HLA alleles with less similar sequences can be reflected less significantly.
[0053] Figure 3b is a table showing the predicted immunogenicity values for the target epitope of HLA alleles similar in sequence to the target HLA allele and the immunogenicity values for the target epitope of the target HLA allele inferred through the weighted sum of these values.
[0054] Referring to Fig. 3b, the distance and weight of HLA alleles with sequences similar to the target HLA allele can be determined to be inversely proportional. For example, if the distance of an HLA allele with sequences similar to the target HLA allele is short, the weight can be determined to be large, and if the distance of an HLA allele with sequences similar to the target HLA allele is long, the weight can be determined to be small. Subsequently, the immunogenicity value of the target epitope for the target HLA allele can be inferred by summing the predicted immunogenicity values of the target epitope for similar HLA alleles by weight.
[0055] The quantified vectors are transformed for each PC axis through PCA transformation, focusing on regions with large sequence differences between HLA alleles, and were able to explain 84.6% of the total variation when up to 9 PC axes were considered. In this embodiment, up to 9 PC axes were considered, but it is not limited thereto.
[0056] PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9 Sum ExplainedVariance 26.0% 19.8% 14.6% 8.7% 5.1% 3.6% 3.0% 2.0% 1.8% 84.6%
[0057] Each HLA allele is plotted as a single point on a 9-dimensional PCA plot to calculate the similarity distance between the target HLA allele and the reference HLA allele using methods such as distance-based (Euclidian distance) and angle-based (cosine similarity). In this embodiment, sequence similarity was calculated based on the Euclidean distance. Clusters can be constructed based on a Euclidean distance of D or less, or N reference HLA alleles similar to the target HLA allele. Since HLA alleles with relatively low similarity are placed in the same cluster, the corresponding HLA allele is reflected less by the summation weighting. Therefore, in this embodiment, N was set to 10 to utilize the immunogenicity prediction values for all alleles that can be calculated by the program among the HLA alleles included in the cluster for a limited number of epitope-HLA allele binding information to predict the immunogenicity of the target HLA allele.
[0058] Immunogenicity estimates for the target epitopes of the target HLA alleles were calculated by weighted summing of predicted values calculated for the target epitopes of each reference HLA allele based on the calculated similarity distances. In the case of sNebula, sequence similarity was measured and weighted summing based on the sum of BLOSUM50 scores for 48 pseudo sequence regions, and in the case of KNN, sequence similarity (distance) information between HLA alleles based on the Euclidean distance of the PCA transformed vector for the entire sequence or the sum of BLOSUM50 scores for 48 pseudo sequence regions was used as the column vector used for imputation to make an equivalent comparison with the method of the present invention and sNebula.
[0059] As parameters used in sNebula or KNN methods, whether to utilize reference HLA allele information (bipartite network neighbors) (n) or to classify a group of neighbors within a certain distance (k) can affect the accuracy of inference. [Table 3] compares the inference performance of the method according to the present invention with that of sNebula and KNN methods using all available HLA allele information, and [Table 4] compares the inference performance based on the parameter that yielded the highest inference performance when evaluating the inference performance of the method according to the present invention, sNebula, and KNN methods while adjusting the parameters. When referring to both [Table 3] and [Table 4], the method according to the present invention, which infers through sequence encoding / PCA transformation / vector distance (N=10), showed the highest correlation with the actual calculated value based on the Pearson Correlation R value, and the correlation between the inference value and the calculated value was statistically significant.
[0060] Method netMHCpan netMHCstabpan netMHCphospan PRIME R Pval R pval R Pval R Pval PCA_dist 0.659 1.1e-27 0.553 2.5e-18 0.286 2.5e-05 0.333 7.2e-07 sNebula 0.602 3.2e-22 0.487 6.1e-14 0.253 2.1e-04 0.235 5.9e-04 KNN full_PCA 0.597 9.4e-22 0.460 1.9e-12 0.261 1.2e-04 0.242 3.9e-04 pseudo_Blosum 0.557 1.5e-18 0.430 6.5e-11 0.214 1.7e-03 0.201 3.3e-03
[0061] Method netMHCpan netMHCstabpan netMHCphospan PRIME R Pval R pval R Pval R Pval PCA_dist 0.659 1.1e-27 0.553 2.5e-18 0.286 2.5e-05 0.357 1.0e-07 sNebula 0.604 2.3e-22 0.487 6.1e-14 0.253 2.1e-04 0.318 2.4e-06 KNN full_PCA 0.598 7.8e-22 0.499 1.2e-14 0.271 6.6e-05 0.257 1.6e-04 pseudo_Blosum 0.557 1.5e-18 0.430 6.5e-11 0.214 1.7e-03 0.202 1.7e-03
[0063] FIG. 4 is a block diagram of a device for predicting the immunogenicity of an epitope for an HLA allele based on sequence similarity according to one embodiment of the present invention.
[0064] Referring to FIG. 4, an apparatus for predicting the immunogenicity of an epitope for an HLA allele based on sequence similarity according to one embodiment of the present invention may include an input / output module (410), a memory (420), and a processor (430).
[0065] The input / output module (410) (or input / output module) may be configured to receive an HLA allele for predicting immunogenicity or to output the predicted immunogenicity of an epitope for the HLA allele. The input / output module (410) may receive an HLA allele for predicting immunogenicity from a user or from another device. Here, the input / output module (410) is shown as a single unit, but it may be configured as a separate device. For example, the input module may be a mouse and / or keyboard, and the output module may be a display such as a monitor or a speaker.
[0066] The memory (420) can store a disclosed HLA allele sequence or HLA allele cluster, etc. The memory (420) can further store a specific program / algorithm / tool, etc. for measuring the immunogenicity of an epitope for an HLA allele.
[0067] The processor (430) can measure the sequence similarity between an HLA allele to predict immunogenicity and other HLA alleles within a cluster containing it. The HLA allele to predict immunogenicity may not be obtained or may not have been learned through a specific program or / and a specific method. The HLA allele cluster (or cluster) may be formed by collecting publicly available HLA alleles and clustering the collected HLA alleles using a sequence clustering algorithm or software. For example, a cluster containing an HLA allele to predict immunogenicity may include all or at least some of the HLA alleles within a certain distance of the HLA allele to predict immunogenicity. As another example, a cluster containing an HLA allele to predict immunogenicity may include a fixed number of HLA alleles in order of proximity to the HLA allele to predict immunogenicity. As yet another example, because the types and frequencies of HLA alleles possessed by race / group differ, the HLA allele cluster may be clustered by further considering at least one of race or group.
[0068] The processor (430) can determine the weights of other HLA alleles within a cluster containing the predicted HLA allele based on the measured sequence similarity. For example, the weight of an HLA allele with high sequence similarity can be determined to be high, and the weight of an HLA allele with low sequence similarity can be determined to be low.
[0069] The processor (430) can determine the immunogenicity of the epitope for HLA alleles within a cluster containing the predicted HLA allele. The immunogenicity of the epitope for HLA alleles within a cluster can be determined through a specific program or / and a specific method. The immunogenicity determined through the specific program or / and a specific method can be expressed as a numerical value.
[0070] The processor (430) can predict the immunogenicity of the epitope for an HLA allele to determine immunogenicity based on the immunogenicity of the epitope for the identified HLA alleles within the cluster and the weights of the determined HLA alleles within the cluster. The processor (430) can predict the immunogenicity of the epitope for an HLA allele (A) using, for example, only some of the HLA alleles within the cluster.
[0072] Foregoing, specific parts of the present invention have been described in detail. It will be apparent to those skilled in the art that such specific descriptions are merely preferred embodiments and do not limit the scope of the invention. Accordingly, the actual scope of the invention is defined by the appended claims and their equivalents.
Claims
Claim 1 A method for predicting the immunogenicity of an epitope for an HLA allele, comprising: a step of measuring sequence similarity between an HLA allele to be predicted for immunogenicity and other HLA alleles within a cluster including the HLA allele to be predicted for immunogenicity; a step of determining weights of other HLA alleles within the cluster based on the measured sequence similarity; a step of confirming the immunogenicity of an epitope for other HLA alleles within the cluster; and a step of predicting the immunogenicity of an epitope for the HLA allele to be predicted for immunogenicity based on the immunogenicity of the epitope for the confirmed HLA alleles and the weights of the determined HLA alleles, wherein the HLA allele to be predicted for immunogenicity is an HLA allele that has not been learned or secured and therefore cannot predict immunogenicity for the epitope through a specific program or specific method, and the cluster is formed by collecting publicly available HLA allele sequences and clustering them based on sequence similarity. Claim 2 delete Claim 3 A method for predicting the immunogenicity of an epitope for an HLA allele according to claim 1, wherein the HLA allele cluster is formed by collecting publicly available HLA alleles and clustering them based on sequence similarity using a sequence clustering algorithm or software. Claim 4 A method for predicting the immunogenicity of an epitope for an HLA allele, wherein, in paragraph 3, the HLA allele cluster is clustered by further considering at least one of race and population. Claim 5 A method for predicting the immunogenicity of an epitope for an HLA allele, wherein the sequence clustering algorithm or software targets the entire length of the HLA sequence or extracts a pseudo-sequence region that forms a core binding with an epitope peptide and clusters by measuring sequence similarity only for the sequence of that region. Claim 6 A method for predicting the immunogenicity of an epitope for an HLA allele, wherein the sequence clustering algorithm or software is a sequence clustering algorithm or software that clusters by comparing the degree of agreement of the sequence itself or clusters HLAs with similar constituent amino acid properties by utilizing physicochemical properties or a BLOSUM matrix. Claim 7 A method for predicting the immunogenicity of an epitope for an HLA allele, wherein the criteria for clustering the collected HLA allele sequences are at least one of the degree of sequence similarity, the distance between vector coordinates (Euclidean distance), the vector angle similarity (cosine similarity), and a predetermined number. Claim 8 A method for predicting the immunogenicity of an epitope for an HLA allele, wherein the step of predicting the immunogenicity of the HLA allele is a step of predicting the immunogenicity of the HLA allele by using only some of the HLA allele clusters based on the weights of the determined HLA allele clusters. Claim 9 A method for predicting the immunogenicity of an epitope for an HLA allele according to claim 1, wherein the cluster containing the HLA allele to be predicted for immunogenicity includes HLA alleles within a certain distance of the HLA allele to be predicted for immunogenicity. Claim 10 A method for predicting the immunogenicity of an epitope for an HLA allele according to claim 1, wherein the cluster containing the HLA allele to be predicted for immunogenicity comprises a number of HLA alleles determined in order of proximity to the HLA allele to be predicted for immunogenicity. Claim 11 A computer-readable recording medium for recording a computer program for executing a method according to any one of paragraphs 1 and 3 through 10. Claim 12 An input / output module that receives an HLA allele to predict immunogenicity and outputs the predicted immunogenicity; a memory storing a publicly available HLA allele sequence, a clustering algorithm, or software; A device for predicting the immunogenicity of an epitope for an HLA allele, comprising a processor that measures sequence similarity between the HLA allele to be predicted for immunogenicity and other HLA alleles within a cluster including the HLA allele to be predicted for immunogenicity, determines weights of other HLA alleles within the cluster based on the measured sequence similarity, confirms the immunogenicity of an epitope for other HLA alleles within the cluster, and predicts the immunogenicity of an epitope for the HLA allele to be predicted for immunogenicity based on the immunogenicity of the epitope for the confirmed HLA alleles and the determined weights of the HLA alleles, wherein the HLA allele to be predicted for immunogenicity is an HLA allele that cannot have its immunogenicity for an epitope predicted through a specific program or specific method because it has not been learned or secured, and the cluster is a cluster formed by collecting publicly available HLA allele sequences and clustering them based on sequence similarity.
