Novel machine learning method allerStat identifies statistically significant allergen-specific patterns in protein sequences

Cutting-edge technologies such as genome editing and synthetic biology allow us to produce novel foods and functional proteins. However, their toxicity and allergenicity must be accurately evaluated. It is known that specific amino acid sequences in proteins make some proteins allergic, but many of these sequences remain uncharacterized. In this study, we introduce a data-driven approach and a machine-learning method to find undiscovered allergen-specific patterns (ASPs) among amino acid sequences. The proposed method enables an exhaustive search for amino acid subsequences whose frequencies are statistically significantly higher in allergenic proteins. As a proof-of-concept, we created a database containing 21,154 proteins of which the presence or absence of allergic reactions are already known and applied the proposed method to the database. The detected ASPs in this proof-of-concept study were consistent with known biological findings, and the allergenicity prediction performance using the detected ASPs was higher than extant approaches, indicating this method may be useful in evaluating the utility of synthetic foods and proteins. Cutting-edge technologies such as genome editing and synthetic biology allow us to produce novel foods and functional proteins. However, their toxicity and allergenicity must be accurately evaluated. It is known that specific amino acid sequences in proteins make some proteins allergic, but many of these sequences remain uncharacterized. In this study, we introduce a data-driven approach and a machine-learning method to find undiscovered allergen-specific patterns (ASPs) among amino acid sequences. The proposed method enables an exhaustive search for amino acid subsequences whose frequencies are statistically significantly higher in allergenic proteins. As a proof-of-concept, we created a database containing 21,154 proteins of which the presence or absence of allergic reactions are already known and applied the proposed method to the database. The detected ASPs in this proof-of-concept study were consistent with known biological findings, and the allergenicity prediction performance using the detected ASPs was higher than extant approaches, indicating this method may be useful in evaluating the utility of synthetic foods and proteins. Food allergies are atopic disorders, which can be classified into immunoglobulin E (IgE)-mediated and non-IgE–mediated disorders. Allergies to cow’s milk, egg, wheat, and peanuts, for examples, are IgE-mediated. Food allergies are typically caused by hypersensitivity of the immune system to specific proteins in foods, which brings about various allergic reactions ranging from itchiness, swelling of the tongue, vomiting, diarrhea, hives, trouble breathing, low blood pressure, and systematic anaphylaxis in severe cases (1Barry Kay A. Allergy and allergic diseases.New Engl. J. Med. 2001; 344: 30-37Crossref PubMed Scopus (934) Google Scholar, 2Kagan R.S. Food allergy: an overview.Environ. Health Perspect. 2003; 111: 223-225Crossref PubMed Scopus (97) Google Scholar). Gupta et al. (3Gupta R.S. Springston E.E. Warrier M.R. Smith B. Kumar R. Pongracic J. et al.The prevalence, severity, and distribution of childhood food allergy in the United States.Pediatrics. 2011; 128: e9-e17Crossref PubMed Scopus (1090) Google Scholar) showed that children under 18 years in the United States yielded an estimated food allergy prevalence of 8% and that approximately 40% of patients with food allergies have experienced a life-threatening allergic reaction. Food antigen-specific IgE antibodies play a pivotal role in most of immediate hypersensitivity to food components. The cross-linking of IgE receptors on mast cells and basophils triggers the release of mediators such as histamines and proteases (4Menon A.K. Holowka D. Webb W.W. Baird B. Cross-linking of receptor-bound ige to aggregates larger than dimers leads to rapid immobilization.J. Cel. Biol. 1986; 102: 541-550Crossref PubMed Scopus (87) Google Scholar). Antigen-specific IgE production from B cells requires for T-cell help, and T-cell–derived cytokine induces the class-switch of B cells into IgE-producing plasma cells (5Mongini P.K. Paul W.E. Metcalf E.S. T cell regulation of immunoglobulin class expression in the antibody response to trinitrophenyl-ficoll. evidence for t cell enhancement of the immunoglobulin class switch.J. Exp. Med. 1982; 155: 884-902Crossref PubMed Scopus (77) Google Scholar). The production of such cytokines from helper T cells is regulated by T-cell receptors, which mediate signaling through the affinity to T-cell epitope peptides presented on human leukocyte antigen (HLA) molecules expressed in professional antigen-presenting cells, and thymic microenvironments formed by medullary thymic epithelial cells (mTECs) enable T-cells lineage commitment by presenting self-antigens (6Klein L. Robey E.A. Hsieh C.-S. Central cd4+ t cell tolerance: deletion versus regulatory t cell differentiation.Nat. Rev. Immunol. 2019; 19: 7-18Crossref PubMed Scopus (122) Google Scholar). Some of the proteins contained in foods are the cause of food allergies. Currently, the allergenicity of proteins has been evaluated by a method based on the homogeneity to known allergenic proteins. In many countries, the method according to the Food and Agriculture Organization–World Health Organization criterion is used, in which proteins having >35% similarity in the 80-aa sliding window of allergen proteins or those identical to six-to-eight contiguous amino acids that are contained in allergen proteins are regarded as allergenic proteins (7FAO/WHOReport of a joint fao/who expert consultation on allergenicity of foods derived from biotechnology.Eval. Allergen. Genet. Modified Foods. 2001; : 12-13Google Scholar). However, novel foods and functional proteins can now be created using genome editing technologies and synthetic biology. The goal of this study is to develop a prediction method that is more reliable even for such novel proteins. Recently, machine learning (ML) has been applied to various research field, and various bioinformatic methods have also been proposed to improve the prediction performance of Food and Agriculture Organization/World Health Organization rules. Most of them are built on whether the protein contains a similar amino acid subsequence to known allergen-specific peptides, which we call allergen-specific patterns (ASPs) (8Ladics G.S. Cressman R.F. Herouet-Guicheney C. Herman R.A. Privalle L. Song P. et al.Bioinformatics and the allergy assessment of agricultural biotechnology products: industry practices and recommendations.Regul. Toxicol. Pharmacol. 2011; 60: 46-53Crossref PubMed Scopus (67) Google Scholar, 9Nicola J.S. Ghantous H.N. Ladics G.S. House R.V. Gendel S.M. Hastings K.L. Workshop proceedings: challenges and opportunities in evaluating protein allergenicity across biotechnology industries.Int. J. Toxicol. 2013; 32: 4-10Crossref PubMed Scopus (6) Google Scholar, 10Stadler M.B. Stadler B.M. Allergenicity prediction by protein sequence.FASEB J. 2003; 17: 1141-1143Crossref PubMed Scopus (179) Google Scholar, 11Riaz T. Ley Hor H. Krishnan A. Tang F. Li K.-B. Weballergen: a web server for predicting allergenic proteins.Bioinformatics. 2005; 21: 2570-2571Crossref PubMed Scopus (23) Google Scholar, 12Li K.-B. P. Krishnan A. allergenic proteins using PubMed Scopus Google Scholar, A. D. a web server for in assessment of protein PubMed Scopus Google Scholar, D. T. of allergenic proteins a of with in and machine PubMed Scopus Google Scholar, L. A. predicting using on allergen PubMed Scopus Google Scholar, prediction of allergenic proteins and of ige PubMed Scopus Google Scholar). these bioinformatic methods are in their to allergen proteins that known we a method to ASPs by data-driven The goal of this study is to ASPs that are for allergenicity from a database of allergen proteins and proteins. an method to statistically reliable ASPs and them to food allergic is and statistically The is the of subsequences in protein amino acid sequences is we to develop a that enables an exhaustive The is a is from a of is to the and of the to is to develop a method to the The of is that these and challenges by P. A. R. of patterns using on and Scopus (179) Google Scholar, an for Scholar, J. J. J. T. using a of the on and for Google Scholar, J. J. B. J. H. et patterns by the Scopus Google Scholar, an for 2001; Scopus Google Scholar, on 2011; Scopus Google Scholar, R. R. and performance on database Google Scholar, P. R. R. of Scholar) and methods D. for 2005; Scopus Google Scholar, H. B.M. and and 2005; : Scholar, H. B.M. et the distribution of Scopus Google Scholar, R. D. in R. PubMed Scopus Google Scholar, B.M. et based on in J. Scopus Google Scholar, R. L. A. L. et 2013; Scopus Google Scholar, R. J. R. for the PubMed Scopus Google Scholar, J. R. R. for and the Google Scholar, for of the on of Scholar, H. for by Scholar, H. R. for by using in Scholar, of by Scopus Google Scholar, and for using on of Scholar). As a proof-of-concept study, we a of 21,154 proteins classified according to whether have been for allergic reactions allergic and by them from applied to the and statistically ASPs the the of that a of the ASPs have which a role in atopic and food allergies B. C. B. prediction of antigen through and of 19: PubMed Scopus Google Scholar) and are consistent with known IgE H. T. T. food and their PubMed Scopus Google Scholar). we develop an allergic prediction method based on the ASPs and that prediction performance of the method is than methods C. a by to in of the on for for Scholar, allergen prediction using PubMed Scopus Google Scholar, a server for in prediction of Scopus Google Scholar). the database for the study in the and the is In this we an method for amino acid subsequences with allergic which we call As a we a protein that contains allergic and proteins in various biological and we applied to the The of the for is in of the an protein of of amino acid biological and the presence or absence of allergic call a protein that or cause allergic as an allergenic protein or a is to patterns as the we the as in in which biological in the is classified into The contains allergen and proteins the contains allergenic proteins and the contains proteins The of these of is amino acid subsequences in a be whether are specific to the or to an allergic reaction. As a we a of 21,154 proteins whose presence or absence of allergic reactions have been already the The in the of is in and the biological of the in the of is in The of allergenic proteins and proteins. The and of the proteins are and are and It is to allergenic and proteins. we on foods that have been for and their proteins were based on protein from and proteins were the thymic a in which the self-antigens are presented that T cells are from the T cell to the T cell play a pivotal role in by of their expression of proteins. such a protein expressed in allergic The and protein were to proteins. this is for prediction performance we to of the that many similar proteins are contained in we we the of of allergenic and for these are by the for amino acids in with the for amino acid presented in is in the of which that is to allergenicity by using those of amino acid sequences and patterns call contiguous amino acid subsequences of various The goal of this study is to find the patterns with allergic reactions as The challenges from the that an of patterns to be we patterns to are amino are In we have to patterns contained in the but even the of patterns in the is as as an of patterns but also The in this is to develop a method for these and The of patterns contained in the by their is in In this study, we call the patterns that the as The of an is is statistically significantly more in allergenic proteins than in is in of the is specific to biological The is to the of the The we that an is as a of a cause of allergic be in proteins. The that patterns specific to a biological be as In this study, the is by whether of the is is in is in a contains allergenic and a more in allergenic proteins can be than specific to the biological the is to whether the contained in a is specific to the biological or the allergic we the patterns as ASPs are contained in the of and of ASPs and the of patterns is is to introduce a to the for of patterns have been in the In methods for patterns are P. A. R. of patterns using on and Scopus (179) Google Scholar, an for Scholar, J. J. J. T. using a of the on and for Google Scholar, J. J. B. J. H. et patterns by the Scopus Google Scholar, an for 2001; Scopus Google Scholar, on 2011; Scopus Google Scholar, R. R. and performance on database Google Scholar, P. R. R. of and method is built on of the methods J. J. B. J. H. et patterns by the Scopus Google Scholar). The of patterns a but also a the methods have been proposed for evaluating the of patterns in for of the on of Scholar, A. H. J. for of the on and Health for Scopus Google Scholar, A. J. for regulation and Scopus Google Scholar, A. J. of A. 2013; PubMed Scopus Google Scholar, L. F. of the most patterns with of the on for Scopus Google Scholar, F. L. and of the on and for Scopus Google Scholar). In this we the of ASPs by the in these to As the of patterns is the be the patterns are in the In a of subsequence patterns is and is by the of the as in In the most can find patterns whose is than a of the that the of the in the is than the of the the of and in a based on can be to statistically whether amino acid subsequence is more in allergenic proteins. the for the of a we a of we to the of a of by are a is to the of In the of is to the of or more the a criterion known as the we we the and the The most method for is the T is the of can that than However, in T is the of patterns and is that the is In are patterns in the those with the than can be statistically for that the of the ASPs can be by with of which to in the and methods have been D. for 2005; Scopus Google Scholar, H. B.M. and and 2005; : Scholar, H. B.M. et the distribution of Scopus Google Scholar, R. D. in R. PubMed Scopus Google Scholar, B.M. et based on in J. Scopus Google Scholar, R. L. A. L. et 2013; Scopus Google Scholar, R. J. R. for the PubMed Scopus Google Scholar, J. R. R. for and the Google Scholar, H. for by Scholar, H. R. for by using in Scholar, of by Scopus Google Scholar, and for using on of Scholar). In this study, we introduce a method for for of by a the method and methods for Scholar) with a we for is that is than the is in the are to a which ASPs by we for the and the patterns from the in method are as findings, to the for must be than the from the with we can the distribution of The is to be the of the the for evaluating the of the ASPs by the method requires the of for be as we introduce a to that the of the we the of the of the patterns in of the in is to the for the patterns which the of the distribution by the based on the The of from methods is that the is that containing patterns the of the distribution in the method are was in A. J. for regulation and Scopus Google in which the goal was to the of In this study, we the method to The ASPs in the are in In and statistically ASPs were the were to be and the of the the and the of the the of the of ASPs in allergenic the distribution of many biological is The ASPs detected by with were to sequences and were to patterns to make for the and the patterns a of amino acids or we of that a role in atopic and food allergies A. J. P. et from Immunol. PubMed Scopus Google Scholar, P. Gupta of and ige in J. Immunol. Google Scholar, and and the a Allergy Immunol. PubMed Scopus Google Scholar). in foods are by antigen-presenting cells such as cells and B are into peptides that can to and presented to T The to such as is as the to antigen-specific IgE antibody The of the and such as was using B. C. B. prediction of antigen through and of 19: PubMed Scopus Google Scholar). the of of or which according to for a is by the of peptides of the of the a is a of that affinity is among the for the of of the with a we the for to for showed for of sequences to are in as The of the of for the and is presented in we the ASPs and known IgE The web server allergen database for food for the similarity with these for which to the IgE from patients with food a search with search As a of were to the sequences of a of with from and an allergen patterns and with epitope sequences in and Some more specific are presented in of were with the known food allergen the of predicting allergic reactions of proteins using the ASPs by prediction from amino acid sequences is for of synthetic foods such as genome is in the of performance by some from the an with the and the to the for performance It is to that is based on the that is the proteins in this study are the biological we we call In proteins in were an is with the proteins in the and the prediction performance is by the to the proteins in the for is a of the presence or absence of an allergic is to the ASPs but also the patterns that are more in we also which were as from in that be significantly more in proteins but may also be in a of allergenic proteins. the to the in the of a is is statistically significantly more in proteins than in allergenic proteins is specific to biological from the we for allergenicity prediction predicting an allergic of a is to the protein is or is to have allergic reaction. Some such as are in that are to the are the of the prediction is in the we a prediction in which a is by the of or in the amino acid In the to a protein and to or the ASPs and in the are patterns that were using the proteins in the by the of a based on the ASPs and the by machine the for the of prediction by and the of the based on the for the the of the ASPs by we the prediction with the methods proposed in are on allergen prediction from amino acid sequences M.B. Stadler B.M. Allergenicity prediction by protein sequence.FASEB J. 2003; 17: 1141-1143Crossref PubMed Scopus (179) Google Scholar, 11Riaz T. Ley Hor H. Krishnan A. Tang F. Li K.-B. Weballergen: a web server for predicting allergenic proteins.Bioinformatics. 2005; 21: 2570-2571Crossref PubMed Scopus (23) Google Scholar, 12Li K.-B. P. Krishnan A. allergenic proteins using PubMed Scopus Google Scholar, A. D. a web server for in assessment of protein PubMed Scopus Google Scholar, D. T. of allergenic proteins a of with in and machine PubMed Scopus Google Scholar, L. A. predicting using on allergen PubMed Scopus Google Scholar, prediction of allergenic proteins and of ige PubMed Scopus Google Scholar, C. a by to in of the on for for Scholar, allergen prediction using PubMed Scopus Google Scholar, a server for in prediction of Scopus Google Scholar, L. Allergenicity prediction by 2013; PubMed Scopus Google Scholar, A. A. C. an method for predicting allergenic proteins and of IgE Scholar). allergen prediction of from amino acid sequences and based on the the goal of the study is to the of the ASPs by we the method with those of and the the ASPs by with in allergenicity prediction allergen prediction using PubMed Scopus Google A. A. C. an method for predicting allergenic proteins and of IgE a server for in prediction of Scopus Google and C. a by to in of the on for for Scholar). for we that can be from amino acid sequences in the The in and are contiguous of amino acid with which is The of is from patterns of the of with or is a method for amino acid subsequences based on a the The of the method by is that is an method and patterns frequencies in allergenic and proteins. from and in that the by are from the of the amino acid The of the method by is that is to which of the amino acid subsequence are with the allergic reaction. the of the patterns in these methods be evaluated. the of the prediction based on the of the for various prediction we evaluated to and for the of of these are presented in and the of the is presented in the is to that the of and the of allergenic and proteins are these are for evaluating prediction for such It can be that the prediction of are than the The on the prediction with the is in The of ASPs based on a database of proteins can be regarded as a for a However, is to that the proteins in the database are which is a in In proteins to the have amino acid sequences that are specific to that and the search is amino acid sequences be of allergen-specific this we the proteins in the database into and for ASPs on whether was or in the allergen prediction must be to make based on amino acid sequences by in which is from the In the protein allergen study, is to into for the of proteins. that of the patterns specific to allergenic proteins by the to of the of the patterns the affinity for of the that can that the sequences the of and in and that patterns many sequences for of the having a than The of the patterns that have to be to their may sequences. the patterns have sequences with or amino acid which have the allergen-specific IgE on food have been the from of which of the of epitope sequences. the epitope sequences were in epitope and and epitope in is in to amino acids to have low of showed with known food allergen are many patterns with among that epitope sequences to the allergen from even epitope was in the The patterns are to T-cell and derived such as those to and antibody in to and were from amino acid in the The a of the Allergenicity is a database were by protein expressed in foods and from It is known that are many patients for these foods, and in them that cause an allergic have been allergen but also have been in such as It is that the presence of is low as with foods that food allergy patients are the prediction performance of on the known in these foods were by the from for Food to the database. to the of by proteins to the for in of the foods, was created by allergen and similar protein from the were by and protein expression C. H. et patterns in human medullary thymic epithelial cells A. 2013; PubMed Scopus Google Scholar, J. et thymic epithelial cells produce Biol. PubMed Scopus Google Scholar). of expression in were with the of t of which a of or was by server which requires amino acid sequences with or The of the allergen-specific sequences was from the to the ASPs having and were using to specific to using similarity was by with E and of The sequences of were from As a for this we ASPs in the sequences and the sequences and as of such that and that the is the we as an The of or more sequences is as the of an for sequences. we a as a an allergen in the such that is in as a subsequence by and is as an of of we the that we those that are in In to find IgE to we the that whether for in IgE epitope sequences and in the database of IgE we and we of the epitope the for as in and the of allergenic and the of and are as and whether is a in the frequencies of patterns allergenic and we the containing in a and the of proteins in and which contains the is The of is by of the under the that and are the of a in a of allergenic proteins on and The is we the that the of can be a by the the is than However, we many we to the the of having or more the the of we to the more that a is the is than the is than for the such that the can be to is in the of The most method is in which the is to be of of is the of to be is known to be the of is In we a the method and methods for Scholar) as a In the are by the or in the is the is to be the among is the of the of to an of is to the for of the for the In we a among patterns as in for the we a to a and the of patterns is to a of the of the patterns in the we can find such a we can the we to the of we a whose is than the we a that the in The in was proposed in method in A. H. J. for of the on and Health for Scopus Google Scholar, A. J. for regulation and Scopus Google in which the to find statistically of In we the in the method for the that the the in can be as in A. H. J. for of the on and Health for Scopus Google Scholar, A. J. for regulation and Scopus Google Scholar). the for as in In to allergic reactions by proteins using ASPs and we the with by ASPs and the of by class with in a that we an of for of and Scholar). that we are of and the of and the of a whose are is that prediction is by the and are to be by is larger than the prediction that the for is the and is a to or in this we the in the prediction performance or by that we for and is by from we to or larger as as the or the is the this is for the by an this for the of is for a of biological we to ASPs by and protein sequences to the of be an or a be and be their allergenicity allergenic or we as the of patterns in