IntroductionRecent years have seen an increase in the numberof biotechnological studies aimed at assessing therole of biologically active peptides derived fromfood raw materials for regulating body functions,maintaining immunological status, and reducing therisk of chronic disease development [1, 2]. Scientistsproved that biologically active peptides demonstrateantimicrobial, hypocholestermic, antihypertensive,antioxidant, antidiabetic, immunomodulatory andotherproperties [3–9]. Peptide of dairy raw materialsis considered one of the most valuable sources forisolating bioactive peptides encoded in its structure[10]. Most biologically active peptides identified in dairyproducts range from 2 to 20 amino acids in length. Thiscorresponds to a molecular weight range of 0.24–2.50 kDa.As the length of the peptide increases, the probabilityof forming secondary structure elements rises, whichresults in steric hindrances to the manifestation ofvarious biological activities. Exposure to proteasesbrings about the release of bioactive peptides from theamino acid sequence of a protein. This exposure takesplace during gastrointestinal digestion, fermentationof milk proteins using proteolytic systems of lacticacid bacteria in the process of ripening, technologicaltreatment of raw materials (homogenization, hightemperature treatment, ultrasound, etc.) and bioconversionof protein raw materials with purifiedpreparations of proteolytic enzymes [11–13].The classical strategy for research of biologicallyactive peptides relies on the unpredictable cleavageof peptide bonds in the protein structure by proteasesin vitro, followed by the purification of hydrolysisproducts and evaluation of their bioactivity in vivo.However, this strategy suffers from a number ofshortcomings, including a high labor intensity anda long process, as well as high financial costs [14].With computer technology and in-depth analyticalresearch methods developing rapidly, integratedproteomic data banks, such as NCBI, BIOPEP,UniProt, PepBank, SwePep, etc. were created. Implementingbioinformatic analysis algorithms on theseplatforms allows the detection of peptide bondsin the protein structure sensitive to proteolyticcleavage, amino acid sequences of proteins and derivedpeptides, their functionality, allergenicity, chelatingability, etc. [15–17].Methods of bioinformatic analysis (in silico) are anauxiliary tool in preliminary studying the biocatalyticconversion of proteins (using “digital twin” models)by different proteases with predicted release ofbiologically active peptides. Since peptides, likeproteins, exhibit a high degree of structure-activityrelationship, the presence and location of certainamino acid residues (biomarkers) can indicate theproperties and potential bioactivity of peptides [18].For example, E.Yu. Agarkova and A.G. Kruchininshowed in their article that redox-active aminoacid residues (C, H, Y, W and M) are an importantstructural descriptor of antioxidant peptides [19].Residues of hydrophobic amino acid enhance theantioxidant properties of peptides in systems containingthe lipid phase. Amino acids with ionogenic groupsin side radicals are responsible for binding metal ionsof variable valence. Thus, predictive modeling ofbiological activities in peptides based on biomarkersreduces the number and duration of experiments toobtain representative data [18]. Bioinformatic analysisintegrated into research developed new strategies fordiscovering bioactive peptides and proving their roleat the organismic level. Most in silico working strategiesare based on a paradigm that selects protein substrateand enzymes to generate bioactive peptides (takinginto account the frequency and release efficiencycriteria), carry out molecular docking, and screenvirtually peptide sequences for further optimizationof biopeptide release from food proteinsubstrates [20, 21].However, the design and generation of biologicallyactive peptides neglect a number of factors. Forexample, the genetic polymorphism of milk proteinsassociated with amino acid mutations in its structure canaffect the type and biological activity of the releasedpeptides [22]. Diversity of the protein matrix of foodraw materials should be considered another importantfactor, as well as their bioavailability for enzymaticcleavage, taking into account the conformationaland intermolecular changes during technologicalprocessing. Considering peptidomics as an integral part offudomics, one should pay special attention to predictingthe sensory characteristics of hydrolysis products,aim to minimize the formation of free amino acids atthe in silico stage, as well as level out the formationof bitterness and non-specific flavor as much aspossible. A key criterion in the development and identificationof biologically active peptides is food safety.That is why a bioinformatic approach to modelingbiologically active peptides should predict such factorsas toxicity and allergenicity of the peptides releasedfrom the protein structure. In terms of technologicalproperties, an important factor is predicting thestability of biologically active peptides during in silicomodeling. Bioinformatics can predict the averagemolecular weight, thermal stability (aliphatic index),solubility (hydropathy index), etc. This enablesassessment of stability for hydrolysis products duringfurther technological processing and storage. Sincebioactive peptides can be completely or partially degradedby digestive proteases in the human gastrointestinaltract and subsequently lose biological activity,49Кручинин А.Г. [и др.] Техника и технология пищевых производств. 2022. Т. 52. № 1 С. 46–57bioinformatic modeling of the resistance of bioactivepeptides to hydrolysis by digestive enzymes is consideredan important part of the final stage. For example, prolinein biologically active peptides increases their resistanceto GI peptidases [19].The foregoing determines the relevance of thestudy objective, i.e. developing a hybrid strategyfor bioinformatic modeling so as to study biologicallyactive peptides of milk protein, taking intoaccount the ranking of key criteria based on highperformanceproteomic database algorithms.Study objects and methodsAnalysis embraced Russian and foreign scientificpublications dealing with the use of bioinformaticdata banks in studying proteins or peptides of foodbiosystems. It was carried out on the main scientometricdatabases RSCI, Scopus and Web of Science.The search query excluded teaching materials, aswell as conference materials and proceedings. Searchdescriptors in article titles, keywords, and abstractsincluded the following words and phrases: food proteins,bioactive peptides, database, bioinformatics, in silico.The depth of analysis for scientific publications waslimited to a 20-year period. This approach allowedus to identify key actualizable databases and formthe fundamental criteria for bioinformatic modelingof targeted hydrolysis of food proteins in order topredict the release of biopeptides from their structures.Results and discussionResultant from the development of principles forthe bioinformatic approach in peptidomics, numerousdatabases were created, including data banks of proteins,as well as enzymes, sensory, allergenic, bioactiveand hypothetically bioactive peptides. In addition tolisting members of each group, the databases containassociated analytical bioinformatics tools. Thanks tothem, one can extract information about the dis-/similarityof given protein structures, their amino acid sequence,theoretical enzymatic cleavage, physicochemicalproperties, chelating ability, proven or predictedfunctionality, allergenicity, toxicity, etc.In a number of studies, scientists used variousbioinformatic resources successfully to createalgorithms and strategies for predicting the isolationof biologically active peptide from food rawmaterials [23–25]. Taking into account the characteristicsof raw materials or the process ofgenerating biologically active peptides, theauthors point out that each individual food object requiresappropriate in silico modeling tools.Analysis and systematization of internationalexperience resulted in development and thoroughdescription of an optimal algorithm for a hybridstrategy of bioinformatic modeling so as to studybiologically active peptides of milk protein. Thestrategy takes into account the most significant criteriathat increase the probability of obtaining peptideswith predictable bioactivity, safety, and acceptablesensory characteristics (Fig. 1).Analyzing the protein profile of dairy raw materials.The fractional composition of raw milk is not constantand depends on paratypical (period of the year,feeding ration, lactation period, animal health, etc.),genotypical (heredity, breed, individual genotype, etc.)and technological (heat treatment, homogenization,membrane processing, etc.) factors [26]. In thisregard, the preliminary proteomic studies requirequalitative and quantitative determination ofprotein fractions for dairy raw materials due totheir instability. To determine the total content ofcasein and serum proteins and to identifyprotein fractions, one needs to use a set of multidirectionaltechniques, such as the Kjeldahl method,one- or two-dimensional gel electrophoresis withisoelectric focusing, high-performance liquidchromatography, etc. In addition, high-performanceliquid chromatography with time-of-flight massspectrometry will assess changes in the peptideprofile in dairy raw materials depending on varioustechnological factors.Thus, complete systematic mapping of proteinsin dairy raw materials, taking into account theconformational and proteomic changes associatedwith the technological features of modern production,seems to be a powerful tool at the initial stage of thebioinformatic modeling strategy.Analyzing the amino acid sequence for a proteintaking into account genetic polymorphism. Thenext stage of the strategy involves obtaining dataon the amino acid sequences of all protein fractionsidentified in the composition of raw milk. Dataon the amino acid sequence, including the proteingene polymorphism (if necessary), its codifiers,molecular weight, and source, can be retrieved frombioinformatic databases and associated tools: NCBI,Uniprot and BIOPEP [27]. These resources areoften used to identify the amino acid sequencesof proteins while studying in silico new bioactivepeptides from animal raw materials and creatingdatabases of sensory peptides [25, 28, 29].However, in silico studies do not take intoaccount information about the genetic variabilityof protein structures.The polymorphism of the gene, encoding theamino acid sequence in the protein structure, playsan essential role in the strategy for bioinformaticmodeling of enzymatic bioconversion of milkproteins. Amino acid mutations result in the random50Kruchinin A.G. et al. Food Processing: Techniques and Technology, 2022, vol. 52, no. 1, pp. 46–57Рисунок 1. Алгоритм гибридной стратегии биоинформатического мод елирования (in silico) для изучениябиологически активных пептидов молочного белкаFigure 1. Hybrid strategy algorithm of bioinformatic in silico modeling to be used in researchon biologically active peptides of milk proteinAnalyzing proteinprofile of milk raw materialsAnalyzing the aminoacid sequence for a proteintaking into account geneticpolymorphismIdentifying bioactiveamino acid sites inthe protein structureScreening the specificityof enzyme preparationsAssessing thebioactivity of peptidesComputer modeling ofthe protein bioconversionAssessing thephysicochemical andtechnological properties ofpeptidesAssessing peptides fortoxicity, allergenicity,free amino acids andsensoricsStability of biopeptidesduring digestionin the GI modelA digital model of a peptidecomplex with predictablebioactivity, safety, andsensory characteristicsSubstrateActive CentreBreaking substrate into subunitsReactionproducts51Кручинин А.Г. [и др.] Техника и технология пищевых производств. 2022. Т. 52. № 1 С. 46–57acid sequence and assessing homology of biofunctionalproperties, as well as identifying precursorproteins [35, 36]. The resultant set containsdata of bioactive peptides with annotated aminoacid sequences included in the studied protein(peptide mapping), their functions, level of bioactivity,and references to primary sources of researchdata. The data set allows one to simplify the processand reduce labor costs of releasing bioactive peptidesfrom complex protein matrices [37, 38]. The targetedhydrolysis will result in the release of not only themaximum possible number of functional peptides,but also those whose bioactivity is not annotated.The bioinformatic tools BLAST NCBI, ExpasySIM Alignment Tool and Uniprot (ALIGN) areused to compare amino acid sequences (alignment)in order to identify protein structures similar inmotifs and functionality [39]. It is worth notingthat working with these resources requires care informulating conclusions. R.A. González-Pech et al.have drawn attention to cases of incorrectinterpretation of the data obtained through thesealgorithms [40].Most other tools used for identifying bioactivepeptides, such as APD, PeptideDB, BioPepDB,etc., operate on the basis of an inverse algorithm[41, 42]. This algorithm focuses on the aminoacid sequences of peptides whose isolation from theprotein requires prior use of resources modellingenzymatic cleavage. This approach forms manyoptions for directing the hydrolysis, since enzymecomplexes or individual enzyme preparationswill have an individual bioinformatic scheme ofcleavage. Processing such a data set implies a timecost, provided that there are no limitations in the numberof enzyme systems. A number of publications onin silico studies of protein microstructuresof collagens, tomato seeds, mung beans, etc.also used this classical algorithm – fromenzymatic cleavage to evaluation of peptideproperties [43–45].Screening the specificity of enzyme preparations.The task of the next stage of the bioinformaticmodeling strategy is to screen the specificity of enzymepreparations taking into account the hydrolysablepeptide bonds at the sites of bioactive peptides. Thebioinformatic tool Expasy Peptide Cutter extractsinformation about the enzymes appropriate for selectedprotein substrates and indicates the hydrolysablepeptide bond between amino acids. Using this information,BIOPEP’s “Batch Processing” provides a list of selectedamino acid sequences and a list of bioactive peptidesincluded in it.Enzymatic screening can also be performed withanother BIOPEP tool, “Find the enzyme for peptidereplacement of single amino acids in the proteinstructure, which affects its properties as well as thebioactivity and degree of peptide release. The effects ofgene polymorphism on the amino acid sequence havebeen noted in a number of studies and constitute a provenfact [30, 31]. Researchers at the University of Limerickstated that the genetic polymorphism of dairy proteinsin raw milk obtained from producing animals of thesame breed affects the types of bioactive peptides itcontains [24]. The direction of hydrolysis can alsodepend on the genetic variation of the protein. Thiseffect has been mentioned in the study of polymorphicvariants of β-casein and their effect on digestion in theGI tract ex vivo [32]. Consequently, when modelingthe targeted hydrolysis of milk protein raw materials,it is necessary to take into account their genotypictraits because they can determine the direction ofhydrolysis and the composition of bioactive sites withinthe protein structure.The fact that dairy plants receive milk from farmsin a bulk milk tank (mixed) poses the main problemfor genetic identification of expressed protein fractionsin raw milk. Milk collected from different cows ischaracterized by heterogeneity of genetic variants ofa certain protein, which complicates its controlledbioconversion. The laboratory of canned milk at theAll-Russian Dairy Research Institute has developed amodern technique for molecular genetic evaluation ofthe ratio of relative shares of the CSN3 gene allelesin mixed dairy products [33]. Based on the proposedtechnique, the authors developed a bioinformatic analysisprogram Calculating the ratio of the relative proportionsof κ-casein alleles in collected milk, available at www.tinyurl.com/allelesprog. Improving this technique andprojecting it onto other biotechnologically relevantprotein fractions will allow integration of this toolinto the strategy of bioinformatic modeling (insilico) from the position of rational processingmilk raw materials for the predicted release ofbiologically active peptides.Identifying bioactive amino acid sites in the proteinstructure. A key step in in silico modeling of hydrolysisis identifing locations of bioactive sites encodedin the amino acid sequences of protein substrates,taking into account genetic polymorphism with theaim of their further targeted release. The evaluationcriterion is the frequency of bioactive sitesoccurrence in the protein structure. Bioactive peptideswithin the amino acid structure of a protein may besearched by its identifier using the bioinformaticdatabase tools MBPDB and BIOPEP [34].Bioinformatic algorithms of these databases are ableto perform a search query in the following variations:searching for bioactive peptides in the structure ofa particular protein; searching for a specific amino52Kruchinin A.G. et al. Food Processing: Techniques and Technology, 2022, vol. 52, no. 1, pp. 46–57peptide database), AHTPDB (antihypertensive peptidedatabase), etc. stand out.Assessing peptides for toxicity, allergenicity,free amino acids and sensorics. Since one ofthe main objectives of biotechnology is to ensurethe safety of isolated substances, a necessarystep consists in testing peptides obtainedby targeted hydrolysis for adverse effects.According to the publications, there areapproximately 170 food allergens that cause IgEmediatedallergic reactions. 90% of these reactionsare caused by food allergens representing 8groups, including milk and dairy products [49, 50].Almost all milk proteins are immunoreactivedue to a large number of antigenic determinants(epitopes) in their amino acid sequences [51, 52].On this basis, a prerequisite for in silico analysisis to predict the residual antigenicity of all hydrolysisproducts. It is possible by means of IUIS andBIOPEP databases containing up-to-date informationon allergenic protein epitopes. In addition tothe search systems of these two bases, there arebioinformatic tools such as Allergenic ProteinSequence Searches and AlgPred2 [53]. They helppredict the allergenicity of isolated peptides and theprotein as a whole by amino acid sequence. Toperform alignment, AlgPred2 is paired with IEDB,which is a database of experimental data on antibodyepitopes studied in the context of infectiousdiseases, allergy, autoimmunity and transplantation,as well as with the NCBI BLAST tool.It is also coupled with the MERCI softwareto identify allergenic sites in the proteinstructure [54].Bioinformatic data on the allergenicity of proteinmicrostructures will allow correcting the hydrolysisprocess by changing the proteolytic system oradding a second hydrolysis step to break downallergenic sites, which is used in practice to reduce foodallergenicity [55].Apart from allergenicity, toxicity of substancesshould be taken into account. It is evaluated usingToxinPred. It is a web server based on a peptidedataset consisting of 1805 toxic peptides obtained fromvarious databases (ATDB, Arachno-Server, Conoserver,DBETH, BTXpred, NTXpred and SwissProt)[56]. There is evidence that certain amino acidresidues, such as Cys, His, Asn, Pro, or the Phe-Lys-Lys, Leu-Lys-Leu, Lys-Lys-Leu-Leu, Lys-Trp-Lys,Cys-Tyr-Cys-Arg sites, are frequently found intoxic peptides, whereas Arg, Leu, Lys, and Ile arethe least common [56, 57]. Bioinformatic tools forpredicting toxicity in silico work on the principleof analyzing amino acid sequence for specificamino acid sites [58]. Current computationalrelease”, where the raw data are bioactive peptidesand the amino acid sequence of the protein fromwhich they are to be extracted. It is important to enterpeptides in FASTA format as follows: “> peptide 1 IPP(amino acid sequence of bioactive peptide)”. Therecan be several peptides, and each must be specifiedwith a new line and a new number. The result ofthe data processing is a list of enzymes suitable fortargeted hydrolysis.Computer modeling of the protein bioconversion.After suitable enzymes are selected in this way, allenzymatic cleavage products can be analyzed inBIOPEP’s section “Enzymes action” by selecting theoption “Enzymes action for your sequence”. This toolfeatures the complete picture of protein hydrolysisinto peptides. Even taking into account the polyenzymesystem. Computer modeling of bioconversionshould be performed on a “digital twin” model of thesubstrate. A digital twin is formed from the analyticaldata on the protein profile of the raw milk used.Bioconversion modeling is carried out for eachprotein fraction, after which the hydrolysis productsare combined and analyzed. The only drawback ofthis scheme is that this tool does not take intoaccount the hydrolysis conditions, namely temperature,duration, substrate-enzyme ratio and pH.This offers the basis for studies to optimize theconditions of enzymatic hydrolysis, takinginto account technological factors in vitro.Assessing the bioactivity of peptides. Aftertargeted hydrolysis on the “digital twin”model of the complex protein matrix of dairy rawmaterials with enzymes selected after screening,all reaction products should be evaluated forbiofunctionality by means of tools. They are listedin “Identification of Bioactive Amino Acid Sitesin Protein Structure”. In addition to the describedbioinformatic resources used to determine the bioactivityof peptides, another tool, Peptide Ranker, is worthmentioning. In the study by S. Nebbia et al., it helpedselect only 10 out of 30 000 prognosticallyformed peptides for further study [35]. This resourceidentifies the biological activity of peptidesaccording to certain structural characteristicson a scale from 0 to 1, in which any peptidescoring above 0.5 is considered biologicallyactive [44, 46]. Using this tool Y. Gu et al.evaluated the effect of different types of cultureson the peptide profile of yogurts. M. Tu et al.studied biologically active peptides derived fromcasein hydrolysis [47, 48]. In addition, thereare a number of narrowly focused databasesthat will help in the targeted search for bioactivity.Among such databases, MilkAMP (antimicrobial53Кручинин А.Г. [и др.] Техника и технология пищевых производств. 2022. Т. 52. № 1 С. 46–57approaches used in toxicology are thoroughlydescribed in studies of antidiabetic, antihypertensive,antioxidant peptides and otherbiological objects for bioinformatic safetyassessments [59–63].For the food industry or pharmaceuticals to continueusing bioactive peptides, it is necessary to predicttheir flavor profile and sensory characteristicsin combination. Sensory characteristics ofbiologically active peptides are another significantdescriptor that bioinformatics tools providefor analysis. The taste profile can be predicteddue to the BIOPEP, which contains a databaseof sensory peptides, as well as the BitterDB,which contains peptides with bitter taste [64].In addition to sensory peptides with bitter,sweet and umami tastes, the abnormal tasteprofile for hydrolysates can be formed due toa high index of free amino acids (FAA) [65].This indicator can be evaluated and correctedduring computer modeling of the targeted proteinbioconversion in silico.Assessing the physicochemical and technologicalproperties of peptides. The amino acid sequencein the structure of peptides obtained as a result ofhydrolysis affects the stability of the system,physicochemical and technological properties.They will affect the application scope for the obtainedcomponents. The bioinformatic tool PepCalc wassuccessfully used in a number of studies to predictphysicochemical properties. It can be used to predictpeptide solubility in water, theoretical molecularweight, isoelectric point, total charge as a functionof pH, extinction coefficient, and instabilityindex [66–68]. The importance of predicting theinstability index, characterizing intramolecularstability, lies in the correlation of this index withthe thermostability of peptides. This is a significantfactor in the technological process (heat treatment)and in the microbiological safety of hydrolysisproducts [69]. Therefore, the instability index can beviewed as one of the criteria for evaluatingthe targeted hydrolysis model or a basis for itspossible adjustment.The Expasy ProtParam and ProtPi tools can alsobe used to predict the instability index, half-life,extinction coefficient, hydropathicity (GRAVY) andsome other characteristics.Stability of biopeptides during digestion in thegastrointestinal model. The structure of biologicallyactive peptides can be destroyed in the gastrointestinaltract by the action of digestive enzymes withcomplete or partial loss of biofunctional properties.Therefore, it is pointless to extract biologicallyactive peptides blindly, without taking into accountdegradation in the GI tract. Evaluating peptidestability during simulated digestion is an importantfinal step in a hybrid strategy of bioinformatic modeling(in silico) for targeted hydrolysis. In silicomodeling of digestion can be accomplished via thebionformatic resources described earlier in “Screeningthe Specificity of Enzyme Preparations”. To simulatedigestion in the gastrointestinal tract, three maindigestive enzymes, produced in the human body,are used: trypsin, chymotrypsin and pancreaticelastase [70].Digital model of a peptide complex. Based onthe sequentially generated algorithm in silico, itseems objectively possible to create a digital modelof the peptide complex. The peptide complex withpredicted bioactivity, safety, and sensory characteristicsmay be an object of subsequent scaling studies in realexperimental conditions.ConclusionBy evaluating the capabilities of multi-directionalbioinformatic analysis methods combined withhigh-performance algorithms of proteomic database,it is possible to combine and integrate them into ahybrid strategy for the bioinformatic modeling (in silico)of hydrolysis for targeted release of stablepeptide complexes with predictable bioactivity,stability, safety and sensory characteristicsfrom complex protein matrices of dairy rawmaterials. In the generated hybrid strategyalgorithm for a bioinformatic modeling, themainemphasis is placed on safety due to excludingthe formationof peptide forms that have a negativeimpact on the functioning of human organs andhuman health in general.The data obtained by bioinformatic modeling(in silico) do not always fully correlate with theexperimental data obtained in vitro and in vivoduring targeted hydrolysis of milk protein and yet thehybrid algorithm presented in this article facilitates the accumulation of the necessary primary datato reduce the time and financial costs of realexperiments.However, despite all the advantages of bioinformaticsand various strategies, in silico remains only apreliminary step in a cascade of studies forbiologically active milk protein peptides due tothe impossibility of predicting the theoreticalenzymatic cleavage under various technologicalconditions (temperature, duration, active acidity,substrate-enzyme ratio). This offers the basis forstudies to optimize the conditions of enzymatichydrolysis, taking into account technological factorsin vitro.54Kruchinin A.G. et al. Food Processing: Techniques and Technology, 2022, vol. 52, no. 1, pp. 46–57Conflict of interestThe authors declare that there is no conflict of interestregarding the publication of this article.