DrushbametricsProject

TenthWinterSymposiumonChemometrics

ModernMethodsofDataAnalysis

Russia,Samara,February29–March4,2016

2

SamaraStateTechnicalUniversityInstituteofChemicalPhysicsRussianChemometricsSociety

TenthWinterSymposiumonChemometrics

ModernMethodsofDataAnalysisTheorganizingcommittee

Co-chairmenYuriZolotovDmitryBykov

SecretaryAnastasiiaMelentevaNataliaRyumina

MembersAndreyBogomolovKimEsbensenSergeyKucheryavskiyFedericoMariniAlexeyPomerantsevOxanaRodionovaSergeyZhilin

Address

SSTU,MolodogvardeyskayaSt.244,443100Samara,Russiahttp://wsc.chemometrics.ru/wsc10 e-mail:wsc10@chemometrics.ru

3

ThanksTheWSC-10organizers andparticipantswish express greatest appreciationtothefollowingconferencesponsorsfortheirvaluableeconomicandfriendlyhelp:

artphotonicsGmbHGlobalmodellingLLCFOSSElectricAmericanelementsCSortcompanyEnSpectrLCC

Finally,wearegratefultoalltheWSC-10attendees,lecturers,accompanyingpersons,andvisitorsfortheirinteresttotheconference.

SeeyouagainatthenextWSC-11conference!

4

Usefulinformation

Conferenceandactivities

Conferencesessionswillbeplacedatthemainbuilding.Russianbanya(sau-na)andcross-countryskiwillbeavailablebyrequest.

Meals

Breakfast,lunchesanddinnersaswellasconferencebanquetwilltakeplaceatthediningroominmainbuilding.

Scores&Loadings

Traditional“ScoresandLoadings”meeting,willagainbeorganizedintheold-fashioned,selfserviceKostromastyleandwilltakeplaceinmainbuilding.

CommunicationThemainthreeRussiancellularnetworks,Beeline,MTSandMegafon,haveaproper coverage around the hotel. WiFi Internet access is available in themainbuilding.

MoneyYoumay exchange EURs andUS dollars toRussian rubles (RUR) in Samarabanks,attherailwaystation,orintheairport.

Excursion

Fridaymorning,therewillbeanexcursiontothefamousStalin'sbunkerfromtheWorldWarII.Thebunkerwasbuiltin1942inninemonthsby800engi-neersand2900workerswhoworkedsecretly.Thebunkerremainedhiddenfromthelocalsuntilitwasdeclassifiedin1991andturnedtoamuseum.

MiscellaneousTheconferenceofficiallanguageisEnglish.Everyoneisencouragedtohavehis/herbadgeattached,bothduringthesym-posiumsessionsandsocialactivities.

5

Schemeof“Polytechnic”campus

UsefulphonenumbersAnastasiiaMelenteva,conferencesecretary +79277454932.(mobile)

6

Monday, February 29, 2016 09:00–12:30 Registration

13:00–14:00 Lunch

14:00–14:30 Opening

Session 1 Process chemometrics Chair: Andrey Bogomolov

14:30–15:30 L1 Kim Esbensen MSPC is not enough for proper Process Mon-itoring and Control — Chemometrics can do better: TOS variog-raphy

15:30–16:00 T1 Leon Rusinov Online diagnosis of incipient faults at techno-logical processes on the basis of nonlinear PCA and linear discri-minant analysis

16:00–16:30 Coffee break

Session 2 Methods developing Chair: Kim Esbensen

16:30–17:00 T2 Oxana Rodionova How to solve authentication problems

17:00–17:30 T3 Yulia Monakhova Hybrid chemometric approaches to in-crease efficiency of classification and data fusion techniques

17:30–18:00 T4 Federico Marini Non-linear modeling revisited: advances in locally linear models

18:00–18:30 T5 Alisa Rudnitskaya Applications of the ANOVA-Simultaneous Component Analysis to the multivariate data with underlying experimental design

18:30–19:00 Free time

19:00–20:00 Dinner

20:00–00:00 Scores & Loadings

7

Tuesday, March 1, 2016 08:00–09:30 Breakfast

Session 3 Methods developing Chair: Riccardo Leardi

09:30–10:30 L2 Jose Manuel Amigo Multi-way analysis. Past, present and future

10:30–11:00 T6 Dmitry Kirsanov An approach to calibration transfer based on latent variable space

11:00–11:30 Coffee break 11:30–13:00 Free time

13:00–14:00 Lunch

Session 4 Curve resolution Chair: Oxana Rodionova

14:00–15:00 L3 Ludovic Duponchel When MCR-ALS pushes back the limits of instrumentation

15:00–15:30 T7 Bruno Debus Quantification of lanthanides in complex mix-tures with X-ray fluorescence: three-point calibration models by correlation constrained MCR-ALS

15:30–16:00 T8 Evgeny Karpushkin Using IR spectroscopy and multivariate curve resolution to elucidate mechanism of heat-induced decom-position of an organic complex

16:00–16:30 Coffee break

Session 5 Software and algorithms Chair: Leon Rusinov

16:30–17:00 T9 Andrey Bogomolov TPT cloud: a full-featured chemometrics software online

17:00–17:30 T10 Vladislav Galyanin Joint variable selection and prepro-cessing optimization approach to the multivariate calibration

17:30–18:00 T11 Alexandr Dokuchaev Software for regression linear equations 18:00–18:30 T12 Sergey Kucheryavskiy Speeding up PCA in R 18:30–19:00 Free time

19:00–20:00 Dinner 20:00–00:00 Scores & Loadings

8

Wednesday, March 2, 2016 08:00–09:30 Breakfast

Session 6 Spectroscopy Chair: Ludovic Duponchel

09:30–10:30 L4 Alexey Pomerantsev Looking through a cover: Diffuse re-flectance spectroscopy of hidden objects

10:30–11:00 T13 Yuri Kalambet Peak integration: finally, how many points?

11:00–11:30 Coffee break

11:30–13:00 Free time

13:00–14:00 Lunch

Session 7 Spectroscopy Chair: Oxana Rodionova

14:00–14:30 T14 Natalia Nechaeva Multivariate data analysis in biomedical spectroscopy: cholinesterase activity determination

14:30–15:00 T15 Janos Elek Method development and validation for the quantitative application of mid-range infrared spectroscopy in GLP environment

15:00–15:30 T16 Anastasiia Melenteva Determination of fat and protein con-tent in milk using optical spectroscopy in the range 400-1100 nm

15:30–16:00 T17 Jordi Cruz Monitoring chemical and sensory parameters of tomato product with near infrared spectroscopy

16:00–16:30 Coffee break

Session 8 Chair:

16:30–18:30 Poster Session

18:30–19:00 Free time

19:00–20:00 Dinner

20:00–00:00 Scores & Loadings

9

Thursday, March 3, 2016 08:00–09:30 Breakfast

Session 9 Applications Chair: Alexey Pomerantsev

09:30–10:30 L5 David Hopkins Using the derivative quotient method in re-gression

10:30–11:00 T18 Andrey Kotenko Multi-criteria stochastic optimization of complex chemical production

11:00–11:30 Coffee break

11:30–13:00 Free time

13:00–14:00 Lunch

Session 10 Applications Chair: Dmitry Kirsanov

14:00–15:00 L6 Riccardo Leardi Application of N-way methods to envi-ronmental data from Genoa

15:00–15:30 T19 Vitaly Panchuk X-ray fluorescence analysis of lanthanide mixtures using partial least squares regression

15:30–16:00 T20 Valeriy Babushkin Application of NIR and MIR Foss solu-tions for food analysis

16:00–16:30 Closing

16:30–17:00 Coffee-break

17:00–19:00 Free time

19:00–00:00 Banquet

10

Friday, March 4, 2016 08:00–09:30 Breakfast

09:30–14:00 Excursion

11

Abstracts

12

L01.MSPCisnotenoughforproperProcessMonitoringandControl—Chemometricscandobetter:TOSvariographyKimH.Esbensena,baKHEConsulting,Copenhagen,DenmarkbACABSresearchgroup,DepartmentofChemistryandBioscience,AalborgUni-versity,campusEsbjerg,Denmark

Process monitoring and control in technology and industry is incompletewithoutfullunderstandingofallsourcesofvariation.ItisnotenoughtobeincommandofMultivariateStatisticalProcessControl(MSPC)iftheprocessda-taareaffectedbysignificanterrors toanunknowndegree.Processdataareaffectedbybothmeasurementerrorsaswellas samplingorPATsensorac-quisitionerrors.Thesecategoriescandominateoveranalyticalerrorsbyfac-tors10-20-50(sic)whenprocesssamplingprinciplesandmethodsforreduc-tionarenottakenintoproperaccount.Itisnotsufficientlyknowninchemo-metricsthatPATsignalacquisitiongivesrisetoidenticalerrortypesasphysi-cal sampleextraction; the latter iswellunderstoodand solutionsabound inthe Theory of Sampling (TOS). This presentation will present the intimateanalogybetweenthepopularPATapproachandmoreconventionalphysicalsampleextraction (which isneeded for calibration). In theprocessandpro-ductioncontextit isessentialtoeliminate,orreducemaximally,allunneces-sarycontributions to theTotalMeasurementUncertainty(MUtotal) inordertobeabletoperformvalidprocessmonitoringandcontrol.Thistaskisstand-ardintheTheoryofSampling(TOS),whichformstheonlyreliablescientificframeworkfromwhichtoseekresolution.Avariographicprocessdescriptionallows quantification of the sum-total effect of all unwanted, unnecessarysampling/sensor acquisition errors (plus the conventional analytical error).Continuousvariographicprocessdataanalysis,withproblem-dependentup-dating, isable to issuewarnings thataparticularprocessmeasurementsys-tem isnot, or is no longer fit-for-purpose, andmust thereforebe improved.Thispresentationcallsforaprocessmonitoringparadigmshift-variographyisamissinglinktoMSPC,whichneedsacompleteprocessmonitoringsystemaugmented by an updatable variogram facility. The example of self-contradictionaryFDAregulatorysamplingdemandsforcriticalpharmaceuti-calmixingprocessesisusedtoillustratetheaboveissuesandclaims.

13

L02.Multi-wayanalysis.Past,presentandfutureJoséManuelAmigoDepartmentofFoodSciences,FacultyofSciences,UniversityofCopenhagen

Multi-waymethodshavebeenappliedinchemistryformanyyearsnow.FromthefirstCANDECOM/PARAFACproposalstothenewestPARALIND,manyal-gorithmsandapplicationshavebeendevelopedandpublished.Nevertheless,therearestillmany issues to faceandaspect to improvewhendealingwiththe application of multi-way methods in the real world. This presentationdoesnotpretend tobeanexhaustive revisionofallmulti-waymethodsandtheiralgorithmic features. Instead, thepresentation isaimedat showing thefundamentalmethods onwhichMulti-way analysis is based on (past), theircurrent application field, benefits and drawbacks (present), and, most im-portantly,theimprovementsthatwouldbeneededtomakeMulti-waymeth-odsmore reliable, fast andadaptable toanydata-driven issue (future).Thisflow-chartinmulti-wayanalysiswillbeconductedbyshowingreal-worldap-plicationsinvariousscientificfields(fermentationmonitoring,counterfeitde-tections in foodstuff, metabolomics and pharmaceutical tablets monitoring)usingdifferentanalyticalmethodologies(fluorescencelandscapes,hyperspec-tralimagingandhyphenatedchromatography).

L03.WhenMCR-ALSpushesbackthelimitsofinstrumentationL.DuponchelLaboratoiredeSpectrochimieInfrarougeetRaman,LilleUniversity,Lille,France

Molecular spectroscopy is an essential tool for characterization of complexphysico-chemical systems. This is mainly explained by the wealth of infor-mationavailableinspectrabutalsobytheopportunityofinsituanalysiswithtimeand/orspaceresolution.Ourinsistenceonseekingthetruthoftenen-couragesustopushthelimitsofinstrumentationsometimeswithoutobtain-inganswers.

Insuchacase,manyscientistsconsiderthatwhatcannotbedonetodaywillbepossible tomorrowwith thedevelopmentofanew instrumentalconcept.Hopefully, thisworkwill show thatwe can domany things right nowwith

14

chemometrics and instrumentation. Three different case studieswill enableus to demonstrate the potential and the adaptability of MCR-ALS method.First,hydrationshellofmoleculeswillbeprobedwithterahertzspectroscopy.Inasecondpart,photochromicmoleculeswillbeexploredwithtime-resolvedspectroscopy.The latestcasewill considermid-infraredspectroscopic imag-ingattheSoleilFrenchSynchrotronFacilityforsingleHeLacancercellchar-acterization.

L04.Lookingthroughacover:DiffusereflectancespectroscopyofhiddenobjectsA.L.Pomerantsev,O.Ye.RodionovaSemenovInstituteofChemicalPhysicsRAS,Moscow,Russia

Investigation of samples’ properties through layers of material of differentthickness isdemanded for theprocesscontrol, forbiochemicalanalysis,andformanyotherapplications.Investigationofobjects,whicharelocatedunderacoverisbasedontheanalysisof spectraacquiredusingtheDiffusereflec-tancenearinfrared(DRNIR)spectroscopy.Eachspectrumisacompositionoftheuseful(target)spectrumandthespectrumofthenuisancelayer.Unfortu-nately,DRNIRisamorecomplicatedtechniquethantheconventionaltrans-missionspectroscopy.Therefore,wecannotsimplysubtract thespectrumofthe cover from the cumulative spectrum.To separate these signals,we sug-gestusinganewphenomenologicalapproach,whichemploystheMultivariateCurve Resolution methods. We consider that the nuisance spectrum of thelayer can be presented as a superposition of three signals ctst(λ) + casa(λ)+csss(λ),which are responsible for transmission, absorption and scattering ofthe light beam inside the layer. Spectra ss(λ) are not affected by the layerdepth h, or the nature of the target. On the contrary, coefficients c dependboth on h, and on the target reflection factor r. Applying the conventionaltransmissionspectroscopyaresearcherisalwaysinterestedtoremovereflec-tionandscatteringeffectsfromthespectra.Thepeculiarityofourapproachisthatwe do not remove the scattering and reflection components, but studythemindetailswiththeaimtousethisknowledgeforthetargetspectrumre-covering.

15

Weapplythisapproachtoaparticularsystem,whichconsistsofseverallay-ersofpolyethylene(PE)filmandtheunderlayerwithknownspectralproper-ties.ToseparatetheinformationoriginatedfromPElayersandwantedtargetwemodifythesystemversus:(1)thenumberofPElayers;(2)thereflectancepropertiesof the target sample.Using theMCRapproachweobtained threepurespectras(λ),aswellasthec-profilesthatreflecttheinputofeachtypeofspectrumthataredescribedbycomputablefunctionsofthePEdepth,h,andtheunderlayerreflectionfactor,r.

This concept is thenutilized toobtainahidden target spectrum in the real-worldsystems.Inparticular,wesucceedinreconstructionofthetargetusingasinglespectrumobtainedthroughseveralPElayers.

L05.UsingtheDerivativeQuotientMethodinRegressionDavidW.HopkinsNIRConsultant,BattleCreek,MI,USA

TheoriginalworkofKarlNorrisandWilliamHruschka inapplyingmultiplelinearregression(MLR)onderivativequotientshasbeenreviewedusingnewsoftwaredevelopedusingMatlab.Themethodstartswithfullspectraforasetofsamplesusedforcalibration,andresultsinoneormoretermsinapredic-tionequation(Eq.1)wherethenumerators(num)anddenominators(den)ofeachtermarethevaluesofderivativesatindividuallyselectedwavelengths.

Ypred=B0+B1der(Wnum1)/der(Wden1)+B2der(Wnum2)/der(Wden2)+...Eq.1

The derivatives are calculated by theGapmethodpioneered byNorris, andeachderivative isoptimizedforthecenterwavelength(W), thewidthof thederivative gap, and thewidthof “boxcar” smoothing, inboth thenumeratorandthedenominatorofeach term. Theoriginalworkhasbeenextendedtoutilizederivativesuptofourthorder.Thenumeratoranddenominatorderiv-ativesdonothavetobeofthesameorder.Themethodisparticularlyeffec-tive for samplesexhibiting spectrawithahigh rangeofbaselineoffsetsandbackgroundslopesduetoparticlesizevariability.

Principlesofdevelopingcalibrationmodelsandvalidating themwillbepre-sented. Formeasuringprotein inwheatsamplesusingNIRreflectionspec-troscopy,apredictionequationusingasingle termwassufficient toachieve

16

optimal prediction. It is interesting that even in a more complex systemmeasuring the hemoglobin concentration inwhole blood samples, the opti-mumperformanceofmodelsisobtainedwithmodelsusingonly2terms.

TheDerivativeQuotientmodelsareveryeasy to implement. It issuggestedthat coupling themodels with current chemometricmethods for qualifyingsamples foranalysisprovidesapowerfulnewtechniqueforroutineanalysisofsamples.Thisregressionmethodappearstobecapableofachievinggreat-eraccuracythancanbeobtainedutilizingthestandardmethodsofpretreat-ingthespectraandregressionusingPCRorPLS,possiblybecausemultiplica-tivespectraleffectsaredirectlycompensatedbythequotientterms,whilethechemicaldifferencesareefficientlycalibratedbytheMLR.

L06.ApplicationofN-waymethodstoenvironmentaldatafromGenoaR.Leardi,E.Martini,E.ScalabriniDepartmentofPharmacy,UniversityofGenoa

Theamountofenvironmentaldataavailablebecomesmoreandmoreimpres-sive.Unfortunately,intheverygreatmajorityofcasestheyareanalyzedonedatum (not evenone variable) at a time.This very crude approach leads todecisions that, in spite of having quite relevant effects on daily life, such asstopping the private car traffic or reducing the temperature in private flatsduringwinter,donothaveastrongeffectonthereductionofpollution.

Obviously, amultivariate data analysiswould lead to amuch better under-standingoftheunderlyingphenomenaandthereforetomuchmoreeffectiveactions.

Thedatasetunderstudyhasbeencollected inGenoaduring6years(2007-2012),bymeasuring3variables (CO,NOandNO2) in5different sitesonceper hour. Thismakes a total of 789.120data (6 years * 365/366days * 24hours*3variables*5sites).

It isobviousthatN-wayPCAis thebestwayofdealingwithsuchacomplexandhighlystructureddataset.Thedatacanberearrangedindifferentways,in order to produce different structures with a different number of modes(three,fourandfive).

17

TheapplicationofseveralN-waymodelsallowedtodetectinaveryclearwaythe effects related to sampling site, sampling year, seasonality, day of theweekandhouroftheday.Someconclusionswerequiteobviousandintotalagreementwith theprevious knowledge, but someother resultswerequitedifferentfromwhatwasgivenforsurebythe“experts”oftheproblem.

Ithasbeennoticed that, inspiteof thehighcomplexityof themathematicalapproach, the results can be shown by very simple and highly informativeplotsthatcanbeunderstoodalsobypeoplewithnochemometricknowledge.

18

T01.Onlinediagnosisofincipientfaultsattechnologicalpro-cessesonthebasisofnonlinearPCAandlineardiscriminantanalysisM.R.Galiascarov,V.V.Kurkina,L.A.RusinovSt.PetersburgStateTechnologicalInstitute(TechnicalUniversity),SaintPeters-burg,Russia

The Fisher linear discriminant analysis (LDA) was widely used in patternrecognition, but only recently it began to be used in diagnostic systems toidentifyfaultsoftechnologicalprocesses.However,themajorityoftechnolog-icalprocessesinthechemicalindustryarenon-linearandLDAworkspoorly.In these cases, the improving of classification can be expected when usingnonlineardiscriminantanalysis(NDA).

NDAsimilarlytothenonlinearprincipalcomponentanalysis(NPCA)suggestsat the firststage themappingof theoriginaldata ina featurespaceof largedimensionandthen,atthesecondstage,theuseofconventionalLDA.Inordertoreducetheamountofcomputationthekernelfunctionsareusedasnonlin-earfunctionsformapping.

However,theuseoftheNDAinon-linemodegivesalowpercentageofcorrectclassifications,especially in thecaseof incipient faultsdue tostrongelonga-tionoftheirclasses.

Thecombinedmethodofdiagnosisisofferedinordertoreducetheseelonga-tions.Thecontinuousmonitoringof theprocess is carriedoutbyNPCAandthe fault identificationbyLDA isstartedonlyafter thedetectingof theiroc-currence. Thus, the identification of a fault (its assignment to a particularclass)isproducedinthevicinityofthepointofdetectionincompliancewiththeminimumdistancebetweenthevectorofthecurrentsampleandthecen-troidofthatclass.Fortheconfidentclassificationthedecisionofthepresenceof theparticular fault is takenafterseveralconsecutive identicaldefinitions.Atthesametime,iftheoperatorforanyreasondoesnottakeanyactiontore-turntheprocesstonormaloperation,theblockingoftheLDAaftersuccessfulclassificationoffaultsisproposed.Itpreventsfurtherblurringofclassesiftheincipientfaultswillcontinuetodevelop.

19

Thesemeasureshaveenabledtoobtainmorecompactclassesandsignificant-ly increase the percentage of correct classifications especially in the case ofincipientfaults.Theefficiencyofthemethodhasbeenstudiedwhilediagnos-ingslowlydevelopingfaultsinthepyrolysisofhydrocarbons.

T02.HowtosolveauthenticationproblemsO.Ye.Rodionova,A.L.PomerantsevSemenovInstituteofChemicalPhysicsRAS,Moscow,Russia

Pattern recognition encloses a variety of differentmethods and techniques.Eachtypeofproblemrequiresanapplicationofrelevantmethods.

Authentication is the process of determining whether an object is, in fact,whatitisdeclaredtobe.Typicalauthenticationproblems,relevantforanalyt-ical chemistry and chemometrics, are counterfeit drug detection, food adul-terationdetection,identificationofillegaladditivesinfuels,andconfirmationof geographical origin of products. Class-modelingmethods also called one-classclassifiersarewellsuitedforsolvingsuchtypeofproblems.Thesemeth-odsdeveloptheacceptanceareaaroundthetargetclass,and,thus,delimitthetargetobjectsfromanyotherobjectsandclasses.

Analysts often substitute authentication task with solving discriminationproblems.Awell constructeddiscriminationmethodwillperfectly classifyanewsampleonlyifthissampleisamemberofoneofthepredefinedclasses.However, incasethenewsampledoesnotbelongtoanyofsuchclasses,thediscriminant analysis is unable to properly define the membership of thesample.Thus,suchtechniquesdoapoorauthenticationjob.

OurconsiderationsareillustratedbyacomparisonoftheresultsprovidedbythePLS-DAandDD-SIMCAinapplicationtoareal-worldexample.

20

T03.HybridchemometricapproachestoincreaseefficiencyofclassificationanddatafusiontechniquesYuliaB.Monakhovaa,b,MonikaHohmannc,d,SvetlanaP.Mushtakovab,NorbertChristophd,HelmutWachterd,BerndDiehlb,UlrikeHolzgrabec,DouglasN.RutledgeebInstituteofChemistry,SaratovStateUniversity,Saratov,RussiaaSpectralServiceAG,,Cologne,GermanycInstituteofPharmacyandFoodChemistry,UniversityofWürzburg,Würzburg,GermanydBavarianHealthandFoodSafetyAuthority,Würzburg,GermanyeUMRIngénierieProcédésAliments,AgroParisTech,Inra,UniversitéParis-Saclay,France

Inchemometricsitiscommontoapplyacascadeofmultivariateapproachesto thedataset under study. For example, PCA and cluster analysis are oftenappliedtodetectoutliersandprovideexploratoryanalysisofthedatabeforetheapplicationofmoreadvancedchemometricapproachessuchasdiscrimi-nantanalysis.Anotherpromisingbutyetunexploredresearchdirectionistoincrease theefficiencyof classical chemometricmethodsbysynergetic com-binationoftheiradvantageswithinanewapproach.

Inthisregard,independentcomponentsanalysis(ICA)asapreprocessingtoolwascombinedwithlineardiscriminantanalysismethods(LDAandFDA).Toillustrate the performance of the ICA/DAmethodology, four representativenuclearmagneticresonance(NMR)datasetsofwinesampleswereused.Theclassificationwasperformedregardinggrapevariety,yearofvintageandgeo-graphical origin. The average increase for ICA/DA in comparison withPCA/DA in the percentage of correct classification varied between 6±1-and8±2%.

Second,anewalgorithmwasproposed for thesupervisedanalysisofmulti-block data structures. It associates the advantages of the ability of commoncomponentsandspecificweightsanalysis (CCSWA)toexplain themaximumofvarianceinthedataandPLS-DAframeworktoshiftresultstowardsthetar-geted property. The robustness of the classification results comparingwitheachsinglemethodology(PLS-DAandCCSWA)wasshownontheexampleof

21

the determination of tomato production type (organic/conventional) basedon the fused 1H NMR, IR and stable isotope data (13C/12C, 15N/14N, and18O/16O).

The use of hybrid chemometric approaches such as ICA/DA andPLSDA/CCSWA resulted in improved classification performance comparingwithsingleapproachesandincreasedefficiencyofchemometricmodelling.

T04.Non-linearmodelingrevisited:advancesinlocallylinearmodelsFedericoMariniDept.ofChemistry,UniversityofRome“LaSapienza”,Rome,Italy

The evolution of the analytical instrumentation, especially through the in-creaseintheapplicationofhighthroughputfingerprintingplatforms,hasledtothepossibilitythatmanysourcesofvariation,otherthantheone(s)of in-terest, affect the instrumental signals, resulting in a non-linear relationshipbetween the dependent and independent blocks or, in the case of patternrecognition,inanon-linearseparabilityofthecategoriesinthefeaturespace.

In such cases, the implementation of non-linearmodeling through a locallylinear approach [1-2] can help tackling with the complex relationships be-tweenthedatablocks tobeanalyzed, reducing theprobabilityofoverfittingbut,atthesametime,allowingahighflexibility,astheextentofnon-linearitycan be tuned by a proper selection of themodel parameters (above all, thenumberofnearestneighborstobeincluded).Inthiscommunication,attentionwillbeposedonLocallyWeighted-PLSapproachoriginallyproposedbyCent-nerandMassart[3]andtoitsclassificationanalog(LocallyWeightedPLS-DA)recently proposed by our group [4]. Indeed, although the concept of locallylinear modeling has been present in the literature for almost thirty years,therearemany issueswhicharestillopen.Accordingly, in thepresentcom-municationkeypointssuchastheevaluationoftheinfluenceoftheweightingand/orschemeadoptedandthepossibilityofinterpretingthemodelthroughthe inspection of the local loadings, or the definition ofwhich could be thebestcross-validationscheme,especiallyinthecaseofhighlystructureddata,willbeaddressed,andsomepossiblesolutionswillbepresented.

22

References

1.W.S.Cleveland,S.J.Devlin,J.Am.Stat.Assoc.1988,83,596-610.

2.T.Naes,T.Isaksson,B.R.Kowalski,Anal.Chem.1990,62,660-673.

3.V.Centner,D.L.Massart,Anal.Chem.1998,70,4206-4211.

4.M.Bevilacqua,F.Marini,Anal.Chim.Acta2014,838,20-30.

T05.ApplicationsoftheANOVA-SimultaneousComponentAnalysistothemultivariatedatawithunderlyingexperimentaldesignAlisaRudnitskayaCESAMandChemistryDepartment,AveiroUniversity,Aveiro,Portugal.

Multivariate data sets with underlying experimental design are becomingmoreandmorecommonwithadventofomicstechnologiessuchasmetabo-lomics,proteomics,etc.Metabolomicsdatasets typicallycomprisehundredsofmetabolitesmeasuredintenthsofsamples,thus,requiringuseofmultivar-iate techniques for theanalysisofeffectsofdesign factors.GeneralizationofAnalysisofVariance(ANOVA)tothemultiplevariables–multivariateANOVAorMANOVAisnotapplicabletothetypicalomicsdatasetsasitbreaksdownduetothesingularityofcovariancematrices.Severalapproachestotheanaly-sisofmultivariatedesigndataweresuggestedintheliterature.MostofthemarebasedonthedecompositionofthedatabyPCAorPARAFAC,whichisfol-lowedbytheanalysisofeffects.Thisapproachallowseffectivelyreducenum-berofthevariablesandthereforenumberoftheparameterstoestimatedandleads to thesimpler,moreparsimoniousandmore interpretablemodels.OnofsuchmethodsANOVA-SimultaneousComponentsAnalysis(ASCA)enablesthe decomposition of the original data matrix into sub-matrices containingvariancerelatedtothedesignfactorsandtheirinteractions[1].PCAisrunoneachof thematrices andanANOVA-likemodel is constructed from thePCAmodels of all effects and interactions. Sub-matrices are defined to have or-thogonalcolumnspacesthereforethecolumnspacesofthesub-modelsfittedon thesematrices are alsomutually orthogonal. This allows interpreting allsub-models independently of each other. Since loadings are calculated for

23

eachsub-model independently,contributionof thevariables toeverysourceofthevariationinthedatamaybeidentified.

Applications of the ASCA to the analysis of the lipidomics, plant and fungalmetabolomicsandelectronictonguedatasetswillbediscussed.

References.

1. A.K. Smilde, J.J. Jansen, H.C.J. Hoefsloot, R.-J.A.N. Lamers, J. van der Greef,M.E.Timmerman,Bioinformatics21(2005)3043.

T06.Anapproachtocalibrationtransferbasedonlatentvaria-blespaceV.V.Panchuk1,D.O.Kirsanov1,2,E.S.Oleneva1,M.M.Khaydukova1,2,A.V.Legin1,21InstituteofChemistry,St.PetersburgStateUniversity,St.Petersburg,Russia2Laboratoryofartificialsensorysystems,ITMOUniversity,St.Petersburg,Russia

Aproblemofcalibrationtransferiswell-knowninanalyticalchemistry.Mostoftheworksaddressingthisproblemarecomingfromspectroscopydomain.The problem appears every timewhen onewants to use calibrationmodeldevelopedforoneanalytical instrument(e.g.nearinfrared(NIR)spectrome-ter)withthedataobtainedbyanotherinstrumentofthesametype(anotherNIR spectrometer). This is especially relevant to multivariate calibrationmodelssincetheytypicallyrequirelargenumberofsamplestoestablishreli-able calibration. This large numbermay be hard/long tomeasure again onanother instrument.Duetothefactthattwospectrometersarenot identicalcopiesandtheymaydifferinspectralshape,sensitivity,operatingconditions,etc. direct applicationof a calibrationmodel to thedata fromanother spec-trometerisnormallynotpossible,sinceitleadstounacceptablegrowthofan-alyticalerrorsinquantificationoftargetsampleparameters.Numerousmeth-odsweredevelopedtoperformcalibrationtransferandmostofthemarere-viewed in thework [1].The calibration transferprocedure canbebasedoncorrectionofregressionmodelparameters(slopeandbias)oritcanemploycorrection (conversion) of the second instrument response. The latter casecanbeconsideredinabroadercontextasacalibrationtransferbetweentwoabsolutelydifferentanalyticalmethods.

24

Inthisworkwesuggesttheprocedurefortransferofcalibrationmodelsbe-tweendifferentanalyticalmethods.Itisbasedontheconversionofanalyticalsignalsfromoneinstrumentintotheformatofanotherinstrument.Thiscon-versionisperformedinlatentvariablespaceasfollows:

X1=T1P1t(1),P2=T1tr+X2tr(2),T2=X2dataP2+ (3),X2cor=T2P1t(4).

FirstwedecomposethedataX1fromthefirstinstrumentusingSVD(singularvaluedecomposition)procedure,thenfromthescorematrixT1wechooseon-lythelinescorrespondingtothesamplesavailableinX2trwhichyieldsT1tr.Us-ingT1trmatrix andX2trwe computeP2 – the loadings for the second instru-ment,whereT1tr+ refers tomatrix pseudo-inverse. Thenwe calculate corre-spondingscorematrixforthedatafromthesecondinstrument(X2data)whichweplantouseforpredictionwithcalibrationmodelfromthefirstinstrumentandfinallywecomputecorrectedresponsefromthesecondinstrument(X2cor)whichcanbefurtherusedwiththemodelfromthefirstinstrument.

The performance of the suggested method was tested with simulated datasetsandwithrealdata. In the lattercasecalibrationmodelsconstructed forquantificationofCo,NiandCuintheirtriplemixturesfromenergy-dispersiveX-rayfluorescencemeasurementsweresuccessfullyemployedforpredictionofthesethreemetalsfromUV-Visspectrometrydata.Moreoverthesuggestedmethodcanbeemployedfortraditionalcalibrationtransfer,e.g.betweentwoNIRinstruments.

Itappearsthatthemethodcanhaveabroadrangeofpossibleapplicationsinanalyticalchemistry.

References:

1.S.D.Brown,TransferofMultivariateCalibrationModelsinComprehensiveChemometrics,Elsevier,vol.3,p.345(2009).

25

T07.QuantificationoflanthanidesincomplexmixtureswithX-rayfluorescence:three-pointcalibrationmodelsbycorrelationconstrainedMCR-ALSB.Debus1,D.O.Kirsanov1,2,V.V.Panchuk1,2,A.A.Goydenko1,V.G.Semenov1,A.Legin1,21InstituteofChemistry,St.PetersburgStateUniversity,St.Petersburg,Russia2LaboratoryofArtificialSensorySystems,ITMOUniversity,St.Petersburg,Rus-sia

TotalreflectionX-rayfluorescence(TXRF)isatechniqueofchoiceforvarioustaskswherehighsensitivityandselectivityarerequired.FromasingleTXRFspectrumbothqualitativeandquantitativeinformationcanbederivedbasedonthepositionandthe intensityof theX-rayfluorescenceradiation,respec-tively.Despite itshigh sensitivity, the interpretationofTXRFspectra canbechallenginginsituationswhereseveralchemicalelementsareanalyzedsimul-taneously.Experimentally,theabsorptionoffluorescenceradiationfromoneelement by other chemical elements of the matrix and the overlapping ofcharacteristic lines fromneighboringelementsmayoccur.One consequenceofsuchso-calledmatrixeffectsisthatlowdetectionlimitcannotbeachievedwhichhinderstheuseofTXRFincertainapplications.

In an attempt to improve the precision in quantification of individual ele-ments in complexmixtures, first-order calibrationmethods have been pro-posed.Recently,partial least-squares(PLS)wasproposedforthequantifica-tionof a six elements lanthanidemixtures fromTXRFspectra [1].Themainadvantage of multivariate methods like PLS compared with classical ap-proachessuchasordinary least-squares is thatmatrixeffectscanbeaccom-modated by the model yielding adequate calibration with better precision.Theonlypre-requisitetoperformPLSisthenumberofsampleswhichshouldbelargeenoughtoselectarelevantcalibrationandtestsets.Thiscanbealim-itationinsituationwheremeasurementcostsarehighoraccesstothemeas-urementset-upislimited.

To overcome this limitation,we propose to use the correlation constrainedMCR-ALSmethodrecentlysuggested in[2].This impliesthe implementationofknownconcentrationvaluesofthetargetanalyte incalibrationsetduring

26

the ALS iteration to reach unique solution. Comparedwith PLS, correlationconstrained MCR-ALS yields similar precision but enables the recovery ofpurespectraofindividuallanthanides.Themostintriguingfeatureisthepos-sibilitytosimultaneouslypredicttheconcentrationofallanalytesfromasetof three-point calibration models with RMSEP values around 1.105 mol.L-1.This promising result indicates that accurate quantification of chemical ele-mentsinTXRFspectroscopycanbeachievedwithaminimalnumberofcali-brationsampleswhereaslargeramountofsamplesisrequiredforPLS.

References

1.D.Kirsanovetal.SpectrochimicaActaPartB113(2015)126–131

2.G.Ahmadietal.ChemometricsandIntelligentLaboratorySystems142142(2015)143–150

T08.UsingIRspectroscopyandmultivariatecurveresolutiontoelucidatemechanismofheat-induceddecompositionofanorganiccomplexEvgenyKarpushkin1,NataliyaGvozdik1,SergeyKucheryavskiy2,AndreyBo-gomolov31MoscowStateUniversity,Moscow,Russia2AalborgStateUniversity,Esbjerg,Denmark3ArtPhotonicsGmbH,Berlin,Germany

Manypracticallyimportantprocessesoccuruponelevatedtemperatures,andthe induced chemical transformations are accompanied by gases evolution:heat-induceddegradationofplastics,conversionofacrylicfibersintocarbonones, and fossil fuels pyrolysis – to name just a few. Generally, the evolvedcompounds structure is relatively similar, and therefore their analysis bymeansofvibrationalspectroscopyisaconvenientapproachtowardselucida-tionoftheunderlyingreactionsmechanismaswellasoptimizationofthepro-cess conditions. Modern analytical instruments provide the opportunity tocarry out simultaneous thermogravimetry/differential scanning calorimetryanalysisandIR/Raman/massspectrometryinvestigationoftheevolvinggas-eousproducts.

27

However,elucidationofthemechanismofthereactionsoccurringuponheat-ingisnotcompletelystraightforward,duetoanumberofexperimentallimita-tions.They include, for instance,poorresolutionof thespectralbands, theiroverlapwith theseof tracesofwaterandcarbondioxide in theatmosphere,the lag between the TGA/DSC signal and the spectra registration, and thespecimeninhomogeneity.

InthiscontributionwepresentanddiscusstheresultsofTGA-DSC-IRstudyofheat-induceddecompositionof tris(acetylacetonato)manganese(III)complex.Usingthisprocessasanexample,wehaveindicatedthecomplicationsoftheevolved gas analysis and demonstrated that they can be partially overcometaking advantage of multivariate curve resolution methods. Moreover, wehaveshownthat theconventionalmethodsbasedonTGAand IRanalysisofthesoliddecompositionproductsproduceincompletedatathatisnotenoughforreliableelucidationofthedecompositionmechanism.Finally,wehavede-velopedaworkflowtoextractthemeaningfulinformationfromthesetoftheevolvedgasspectra.Theelaboratedapproachisfairlygeneralandapplicablefor studyofavarietyof theprocessesaccompaniedbyevolutionofgaseousproducts.

T09.TPTcloud:afull-featuredchemometricssoftwareonlineAndreyBogomolov1,2,3,VladislavGalyanin1,VadimBelikov1,ValeriaBelikova11SamaraStateTechnicalUniversity,Samara,Russia2artphotonicsGmbH,Berlin,Germany3GlobalModelling,Aalen,Germany

Chemometricsisoneofthemodernscientificdisciplinesthatcannotbeimag-ined without software. TPT cloud presented here is a new chemometricstoolboxthathasbeenconceivedandabovodevelopedonlineusingup-to-datewebtechnologies.

Themainadvantagesofhavingchemometrics“onacloud”arerelatedtothecentralhostingofthedataandmodelsthatfacilitatestheiruseandshare.Thedata can be easily uploaded from a number of spectroscopic and general-purpose formats and analyzed using a conventional web browser, inde-pendentlyonthehardwareplatformandoperatingsystem.Thedataanalysisworkisthereforeautomaticallybacked-upandsynchronized.Themodelcen-

28

tral location enables its training using multiple data sources that can bespreadallovertheworld,i.e.foraglobalmodelling.Powerfulcomputationalresources that can be involved into the remote computing are necessary tosolve modern “big data” problems. In the cloud chemometrics word heavyfilesarereplacedbyweightlesslinks;thedataandmodelscanbeshared,dis-tributed and reported without cloning. This opens new capabilities for theteamanalysis,corporateworkandscientificpublishing.

TPTcloudisarapidlydevelopingchemometricsplatformthatisnotlimitedtowidelyspreadmultivariateregressionandclassificationtools.Theuniversity-basedsoftware teamdevelopsand implements theirownnovelmultivariatealgorithms.Oneof them is an intensivemodel optimization approachbasedongeneralized interval selectionwithsimultaneouspreprocessingoptimiza-tion.

Chemometrics software is sophisticatedandshouldbecreatedby scientists,butitshouldnotbeintendedforscientistsonly.Thefinalproductofmultivar-iatedataanalysis is amodel that is aimedat solvingvariouspracticalprob-lemsbydifferentpeople,whomaybenotawareoffactorspaces.Connectingspecialistsanduserscanbeeffectivelyperformed in the internetand this isoneofthemainmissionsoftptcloud.com.

T10.JointvariableselectionandpreprocessingoptimizationapproachtothemultivariatecalibrationVladislavGalyanin1,AndreyBogomolov1,2,31SamaraStateTechnicalUniversity,Samara,Russia2artphotonicsGmbH,Berlin,Germany3GlobalModelling,Aalen,Germany

Variable selection is a one of the most common and important treatmentsprecedingthemultivariatecalibration,particularly,ofspectraldata.Irrelevantandnoisydatapartsarecommonlyremovedfromthecalibrationsetinordertosimplifymultivariatemodelsandimprovetheirperformance.

Nowadays,datascientistsusuallydetermineanoptimalpreprocessingmeth-od andperformvariable selection separately, as two independent data pre-treatment steps preceding the calibration. In general, however, the prepro-

29

cessing method, the optimal variables and even the model complexity (thenumberoflatentvariables,LVs)arerelatedtoeachother.Sequentialoptimi-zationmayresultintheselectionofvariables,whichareonlyoptimalforthechosenpreprocessingmethodforagivennumberofLVs.

Variableselectionisusuallyperformedusingthebinaryvariablecoding,asinconventional genetic algorithm (GA) or using a sequentialway, like intervalpartialleast-squares(iPLS)regression.ConventionalGAoptimizationwiththebinarycoding leads toanexcessivecomplexityof theoptimizationproblem,andsequentialintervalselectioncannotguaranteefindingtheglobaloptimalsolution.

Inthepresentworkageneralizedoptimizationapproachforsimultaneousde-terminationofanoptimalpreprocessingmethod,variableintervals(includingtheirwidths)andthenumberofLVsispresented.Theproposedapproachis“time-dependent” but well generalized, has lower optimization complexitythancommonvariableselectionalgorithmsandconvergestoaglobaloptimalsolutioninmanycases.

Here,GAhasbeenusedasaparametricoptimizationroutineinordertoob-tain optimal values of the merit function. Various calibration statistics areconsideredforquantificationanddiscriminationcases.Problemsofchromo-somecoding,geneticalgorithmadaptationandconstraints,intervalwidthandspectral resolution,performance issuesandmerit functionselectionaredis-cussed.

Codefor Intervalselectionwaswritten inGNUOctave/MATLABandused inTPT cloud software (www.tptcloud.com) for online optimization of variableselection“onthecloud”.Allm-filesaredesignedinawaytobeoptionallyusedas an independent standalone GNU Octave/MATLAB toolbox. The software(IntervalSelectiontoolbox)hasbeenregisteredinRussianFederalServiceonIntellectualProperties.

30

T11.SoftwareforregressionlinearequationsA.V.Dokuchaev,A.A.KotenkoSamaraStateTechnicalUniversity,Samara,Russia

Prediction bymeans ofmultivariate regression analysis allows to solve theproblemofmulti-criteriaoptimizationofroadbitumenproductionwithasetofcontradictorycharacteristicsofmodernstandards.Mathematical formula-tionoftheproblemisdescribedin[1].Whiledeterministicproblemisunsolv-ablebecauseoftheopposingclaimsofsomeobjectivefunctions,thestandardrangesofproduct’scharacteristicsofcomplexchemicalproductionallowustofindtheconfidenceintervalstocontrolthetechnologicalparameters.Asare-sult, one can determine the parameters of the production process and thecompositionoftherawmaterialtoproducebitumenwithspecifiedcharacter-istics.Conflictingcriteriaformulti-criteriaoptimizationcanbedescribedasasystemofinterdependentregressionequations.Thenthealgebraicinsolvabil-ityispossibletoovercomebyindicationoftheconfidenceintervalsforthepa-rametersoftechnologytoensurethereleaseofnon-defectiveproduct.

Thesolutionisimplementedinthesoftwarepackage[2].Ittakesintoaccounttheexperimentaldataoflargesamplesandsimulatesvariousformsofinter-connectionsbetweenthecontrolregressorsandtheobjective functions.Thedevelopedsoftwarepackageallowstoquicklycarryoutlargeseriesofexper-imentsandanalyzetherelevantdata.

Wepresent thedetails of themathematicalproblemand the featuresofde-signedsoftware forsubmittingofa largenumberofcontrolparametersandcomplexeffectonthestandardizedcharacteristicsofproduct.

References

1.KotenkoA.P.,SharapovaE.A.BenginaP.M.Multi-criteriaoptimizationusingsystems of linear regression equations // International scientific-technicalconference"Advancedinformationtechnologies",28-30April2015-Samara:SamaraPublishingScienceCentre,2015-P.276-277.(InRussian.)

2.DokuchaevA.V.,KotenkoA.A.Thesoftwarepackageformulti-criteriaop-timization of production systems using regression equations. ROSPATENT.Applicationforstateregistrationofcomputerprograms№2015660647from09.11.2015.(InRussian.)

31

T12.SpeedingupPCAinRSergeyKucheryavskiyDepartmentofChemistryandBioscience,AalborgUniversity,Esbjerg,Denmark

Principalcomponentanalysis(PCA)isawidespreadmethodformultivariatedataexploration. Inmanycases,PCAisthefirsttoolananalystusestogetafeelingofthedatastructure.Analysisofhyperspectralimagesisnotanexcep-tionhereandmanyimageanalysispackagesprovide,amongothertools,thepossibilityforexploringimageswithPCA.

Mostofmultivariatedatasetsusedinchemometricshaveanumberofvaria-bles(significantly) largerthananumberofobjects(observations)andmanyalgorithmsareoptimizedforthiscase.WhenPCAisappliedtoahyperspectralimage,eachpixeloftheimageisanindividualobservation(spectrum).Nowa-daysmany laboratoryhyperspectralcamerascanobtainan imagewithhun-dreds of thousands of pixels. In remote sensing the images can easily haveresolutionup toseveralmillionpixels.Workingwithsuch imagescaneasilylead tomemoryandcomputational issues if standardalgorithms forestima-tionofprincipalcomponentsareemployed.

Inthepresentworkauthorisgoingtosharehisexperienceofcreatinganin-teractivetoolforexploratoryanalysisofrelativelylargehyperspectralimageswithPCAusingRand JavaScriptbaseduser interface.Themaingoalwas tomake the implementation of PCA decomposition as fast as possiblewithoutsubstitutingstandardRtools(e.g.libraryforlinearalgebraoperations),sothepackageiseasilyavailablefornon-experiencedusers.ForthesamereasonitwasdecidedtowriteallcodeinRwithoutusingC/C++atthecurrentstage.

ThreePCAstate-of-artalgorithmshavebeenconsidered—singularvaluede-composition (SVD), eigenvectors of variance-covariance matrix and non-iterative partial least squares (NIPALS). Each algorithmwas investigated inorder to find the steps critical for working with datasets containing hugenumberofrows.Thestepswerethenoptimized(bothalgorithmicallyaswellasanRimplementation)toreducethecomputationaltime.Theresultsofop-timizationwillbediscussed.

32

T13.Peakintegration:finally,howmanypoints?KalambetY.1,KozminY.2,SamokhinA.31AmpersandLtd.,Moscow,Russia2InstituteofBioorganicchemistryRAS,Moscow,Russia3ChemistryDepartment,MoscowStateUniversity,Moscow,Russia

Accuracyofpeakintegrationalgorithmsforthecaseofstand-alonepeak-wisefunctionshapeandconstantdata rate is revisited. It is shown, that the sim-plest rectangle rule gives theunbiasedestimateof the truepeakarea, ifweconsiderframestartpositionasarandomvariable.Mostotherformulas,suchasSimpson’srule,eitherprovidebiasedareaforperiodicnoiseortheirareaisequivalent to thatobtainedbyrectanglerule.Numberofpointsrequired forpeak description with sufficient accuracy depends on the type of measure-ment (instantaneous or integrating) and on parameter under consideration(area,height,width,asymmetry,baselinewidth,…).Mostimportantfactorforareaandheightistheinfluenceofproperbaselinesubtraction.Adaptivenoisefilteringsignificantlyimprovesbaselinedetectionandprovidesbenefittoes-timatesmadewiththehelpoffiltereddata.Integrationrulesinfluenceareainthe case of unseparated peaks only, so that adjacent peaks exchange smallpartsofareas,butinthiscase,areaerrorscannotbecompensatedbysuchex-change.

Italsohastobenoted,thatintegrationparameters,suchasareaorheight,arenotuseddirectly; they firstareusedtoconstructcalibrationcurveandthenquantificationismade.Manyalgorithmicdistortionsarecompensatedonthisstep.

Thebestresultscanbeobtainedbymodellingpeaksaccordingtotheirshape,knowninadvance(e.g.GaussianorEMG).Themoreisknownabouttheshapeofpeaksinadvance,thebettertheaccuracyofapproximation.

33

T14.Multivariatedataanalysisinbiomedicalspectroscopy:cholinesteraseactivitydeterminationNechaevaN.L.,KurochkinI.N.Chemicalenzymologydepartment,FacultyofChemistry,LomonosovMoscowStateUniversity,Russia

Surface-enhanced Raman spectroscopy is a modern powerful tool for theanalysisofcomplexbiologicalsamples.Theuseofsilverpastes isoneof themost common SERS applications on the practice. In the currentwork silverscreen-printedSERS-substrateswereused.Thisdevelopedtechniqueisinex-pensive,easy-to-useanddonotrequiredspecialequipment.TheapplicationoftheSERS-effectisexpedientinthecaseoftheanalysisofcomplexbiologicalobjects,forexamplehumanbloodenzymes.Acetylcholinesteraseandbutyryl-cholinesterase catalyze the hydrolysis of many physiologically importantcompoundssuchasneurotransmitteracetylcholine,caffeine,andcocaine.Theactivityoftheseenzymesinthebloodmaybeloweredduetopoisoningbyor-ganophosphorus compounds, pesticides, and also due to various diseases.Therefore it is very important to measure cholinesterase activity in bloodwith high accuracy to assess toxicity and for diagnosis of diseases at earlystages.ThebestknownmethodfordeterminingtheactivityofAChEandBChEis to measure butyrylthiocholine (BTCh) and thiocholine (TCh) – substrateandproduct of the enzymatic reaction, respectively.Weused a silver print-screensurfaceastheSERS-activestructurefortheanalysisofBTChandTCh,withsubsequentdeterminationofenzymeactivity.

Ramanspectraarealargeamountofdata,whichincludesinformationaboutthecompositionofthesubstance,aswellasvariousnoises.Toassessthecon-tributionofeachpeakinthespectrumandmakequantitativemeasurementsofthespectrumisaverydifficulttaskduetothesheeramountofinformationobtained fromthespectra.Forstatisticallysignificantallocationofcontribu-tionsandforaccuratequantitativeanalysisweusedmultiparametricstatisti-cal methods. In our work we used the principal component analysis (PCA)methodfortheseparationofspectraofsubstrateandproductoftheenzymat-icreaction,andthemethodofPLStoconstructacalibrationcurveandfurtherdefinetheunknownsamples.Ellman’sassaywasusedasreferencemethodtovalidateenzymaticactivityvalues,measuredbyRamanspectroscopy.

34

T15.Methoddevelopmentandvalidationforthequantitativeapplicationofmid-rangeinfraredspectroscopyinGLPenvi-ronmentJánosEleka,b,AngélaSimonb,JánosTörökb,AngélaCsontosb,CsabaBallaibaSciencePortKft,Debrecen,Hungary.bFumoprepKft,Budapest,Hungary

The main analytical role of the mid-range infrared spectroscopy was thestructural investigation of organic compounds for decades until the cheap,versatile and easily interpretablemass spectroscopicmethods slowly occu-piedthearea.NowadaysnewperspectivesareopenfortheMIRusersduetothemultipliedcomputingcapacityofthepersonalcomputers.Itdoesnotonlymean stronger hardware but the development of chemometric data evalua-tionmethodsandsoftware.Nevertheless,thankstothenewsampleintroduc-tiontechniquesthatprovidegoodrepeatabilitywithminorornosamplepre-treatment.Thesignificantlyquickermethoddevelopment, theshorteranaly-sistime,thesmallercostofchemicalsandlabor-intensiveworkphasescom-paredtotheconventionalchromatographicmethodsaretheotherreasonsofthis“renaissance”.

ThelaboratoriesoperatinginGLP/GMPenvironmentmustfulfillrigorousre-quirements.Theanalyticalperformancemustbedemonstratedforanymeth-odsappliedinquantitativedeterminations.Dependingontheapplicationfielddifferentvalidationproceduresexist(suchasFDAorICH1guidelines)buttheyare almost exclusively set up for univariate— mainly chromatographic—methods. As probably the only exemption, there exist methodology for thevalidationofnearinfraredsystems2,butduetothedifferentnatureofNIRandMIRnotalltheparameterscorrespondtoeachother.

In our presentation the development and the validation of a Fourier-transformmid-rangeinfraredmethodwillbedemonstratedwithspecialfocusontheassayspecificityparameter.

References

1. ICHQuality Guidelines Q2A, Text on Validation of Analytical Procedures;Q2B, Validation of Analytical Procedures: Methodology(http://www.ifpma.org/ich5.html)(1996).

35

2.HandbookofVibrationalSpectroscopy, JohnWiley&SonsLtd,Chichester,2002

T16.Determinationoffatandproteincontentinmilkusingop-ticalspectroscopyintherange400–1100nmAnastasiiaMelenteva1,AndreyBogomolov1,2,31SamaraStateTechnicalUniversity,Samara,Russia2artphotonicsGmbH,Berlin,Germany3GlobalModelling,Aalen,Germany

The rapidgrowthof the food industry requiresnewsolutions in thequalitycontrolofprocessesandproducts.Thedevelopmentofnewmethodsenablingquickandaccuratedeterminationoftheanalyzedcomponentsisanimportanttaskespeciallyformassconsumptionproductssuchasmilk.Duetothepres-enceoffatglobulesandproteinmicellesmilkhaspronouncedlight-scatteringproperties. Economically attractive visible and short-wave near infrared(Vis/SW-NIR)spectroscopycanbesuccessfullyusedforthequantitativeanal-ysisofmilk. Inthepresentworkanewscatter-basedapproachtothedeter-minationoffatandproteincontentinmilkhasbeendeveloped.Thedifferenceinscatteringspectralpatternsbyfatandproteinparticlescanbeutilizedfortheiraccuratequantitativedeterminationusingmultivariatemodellingmeth-ods,i.e.partialleastsquares(PLS)regressionanalysis.

In the first step the dependence of diffuse transmittance spectra in theVis/SW-NIR region has been systematically investigated using gradual ho-mogenizationofrawmilksamples.Thehomogenizationresultsinsignificantspectral changes,whichwereexplained in termsof the representative layertheoryandotherscattering theories. The fundamental feasibilityof scatter-based quantitative analysis ofmilk fat and protein content using their low-selectivity diffuse transmission spectra has been proved in a series of de-signedexperiments.Thisnewanalyticaltechniqueisresistanttoanessentialvariability of fat globule sizes thatmay occur in rawmilk samples. The ap-plicabilityoftheapproachtothehomogenizedmilkanalysishasbeentestedona largesetofdifferent samples,whichareavailable in tradenetworks inRussiaandGermany.Finally,theglobal(i.e.resistanttoseasonal,genetic,re-gional and othermilk variations)models for determination of fat and total

36

protein content in rawmilkbasedonhistorical spectroscopicdata collectedduringayearhasbeendeveloped.Inwasalsoshownthatfull-rangespectros-copycanbereplacedbyasetoflight-emittingdiodes(LEDs)asalightsourceandconventionaldigitalcameraasadetector.Thissimplifiedtechnologyhasshownanacceptabledeterminationaccuracyoffatandproteincontentinrawnaturalmilk.

Thepresentedresultshavesignificantpracticalvalueandthedevelopedap-proachcanbeputintothebasisofdifferentdevicessuchasportableinexpen-siveanalyzersoranalyticalsystemsforin-linemonitoring.

T17.MonitoringchemicalandsensoryparametersoftomatoproductwithnearinfraredspectroscopyD.Suna,J.Cruzb,J.Casalsd,J.Sabatec,J.Simóc,ManelAlcala,aDepartamentdeQuímica,FacultatdeCiències,UniversitatAutònomadeBarce-lona,SpainbEscolaUniversitàriaSalesianadeSarrià,Barcelona,SpaincEscolaSuperiord’AgriculturadeBarcelona,UniversitatPolitècnicadeCatalu-nya,SpaindFundacióMiquelAgustí,Sabadell(Barcelona),Spain

Tomatoisoneofthemostimportantcropsintheworldanditsqualitywillaf-fect theheathofpeopleallroundtheworld.Thenear infraredspectroscopy(NIRS)wasadoptedtomonitortheCriticalQualityAttributes(CQAs)oftoma-to,suchasthesolublesolids,drymattercontent,concentrationofglucoseandfructose, sweetness, taste intensity, aroma intensity, mealiness, acidity,crunchiness, juiciness and explosiveness. Quantitative models were con-structed by partial least squares regression (PLSR). The validation modelsshowedthattheNIRShadagoodcapabilityforthequalitycontroloftomatoproductbothinchemicalandsensoryparameters.Forthechemicalparame-ters, rootmeansquareerrorofprediction(RMSEP)was lower than0.5.Forsensory parameters, RMSEP was from 0.2063 to 1.0445 which was lowenoughforthedeterminationofsensoryparametersinfactory.TheNIRSpro-videsusasimpleandrapidtooltorealizetheapplicationofprocessanalysistechnology(PAT)intomatomanufacturing.

37

T18.Multi-criteriastochasticoptimizationofcomplexchemicalproductionA.P.Kotenko,D.A.PsheninaSamaraStateTechnicalUniversity,Samara,Russia

Theproductionofroadbitumenisanexampleoftheproblemofmulti-criteriaoptimization of petrochemical production. Conflicting demands of modernstandards for these products make it difficult to control the parameters ofcomplextechnologicalprocesses.Inaddition,thevariablecompositionofrawmaterials coming fromdifferent fieldsofoil, doesnot allow to fine-tune theproductionprocessby themethodsof feedback control theory.Under theseconditions,itgivesgoodresultsoftheforecastingcharacteristicsoftheprod-uctbymeansofmultivariateregressionanalysisofthesystemsofequationsdescribingsomestandardizedparameters[1,2].Whiledeterministicformula-tionoftheproblemisunsolvedbecauseoftheopposingclaimsofsomeobjec-tive functions, acceptable standards of performance ranges of properties ofproductallowustofindtheconfidenceintervalsofthecontroltechnologicalparameters.

Theproposedmathematicalmodelisimplementedbythesetofprogramsfordata analysis of large experimental sampleswith a variety of analytical de-pendenceofcontrolcovariatesandobjectivefunctions.Itwasinvestigatedtheinfluenceofpartialmanagingregressorsonthespreadofmonitoredparame-ters derived road bitumen. Among the technological parameters (tempera-ture, warm-up time) and the composition of the incoming raw materials,markedsignificantandnon-significantregressors.

References

1.KotenkoA.P.,BukarenkoM.B.Thegeometryoflinearregressionequations// Bulletin of Samara Scientific Center of the Russian Academy of Sciences,t.15,№6(3),2013-P.820-823.(InRussian.)

2.KotenkoA.P.,SharapovaE.A.BenginaP.M.Multi-criteriaoptimizationusingsystems of linear regression equations // International scientific-technicalconference"Advancedinformationtechnologies",28-30April2015-Samara:SamaraPublishingScienceCentre,2015-P.276-277.(InRussian.)

38

T19.X-rayfluorescenceanalysisoflanthanidemixturesusingpartialleastsquaresregressionV.V.Panchuk1,2,D.O.Kirsanov1,3,A.A.Goydenko1,M.M.Khaydukova1,3,S.M.Irkaev2,А.V.Legin1,3,V.G.Semenov1,21InstituteofChemistry,St.PetersburgStateUniversity,St.Petersburg,Russia2InstituteforAnalyticalInstrumentationRAS,St.Petersburg,Russia3Laboratoryofartificialsensorysystems,ITMOUniversity,St.Petersburg,Russia

X-rayfluorescence(XRF)isaveryconvenienttoolofexpressmulticomponentanalysis and it does not require complex sample pretreatment procedures.Neverthelesstherearecertainissueshinderingwideacceptanceofthemeth-od in analytical practice.These issues are considerable samplematrix influ-enceandratherhighdetectionlimits.Duringsimultaneousanalysisofmetalsinsolutionsthesematrixeffectsaremainlytwofold:1)absorptionoffluores-cenceradiationfromanalyzedelementbyotherchemicalelementsofthema-trix;2)overlappingofcharacteristiclinesfromanalyzedelements.Whilema-trixeffectscanbeeasily taken intoaccount inaqueoussolutionbychoosinganappropriatestandardsamples,thesensitivityisstillinsufficientforcertainanalyticaltasks.Apossiblewaytocircumventtheseissuescanbetheapplica-tionofappropriatemultivariatedataanalysistechniques.

There are certain analytical tasks requiring simultaneous quantification ofseveral lanthanides in complex mixtures, e.g. in spent nuclear fuel repro-cessingwhere lanthanides are fission products thatmust be removed fromthemedia;duringprocessingofmonazite—animportantoreforrareearthelementsandduringtechnologicalmonitoringofvariousindustrialsolutions.

To address this task we employed two different X-ray fluorescence tech-niques: energy-dispersive (EDX) and total reflection (TXRF). Comparison ofthis techniques (EDXandTXRF) in simultaneousquantificationof sixneigh-boring lanthanides in complexwater solutionswas performedwith two re-gressionmethods(OLSandPLS).Dueto lowsignal/noiseratioandoverlap-pingsignals inEDXmeasurementsOLSfails toprovideforreasonablepreci-sion in lanthanide determination especially in low concentrations. PLS pro-cessingofthesedataallowedforsignificantimprovementofaccuracy.Moreo-ver,simpleXRFmethodsuchasEDXcanprovideforthesameprecisionasa

39

moresophisticatedTXRF,when themeasurement's resultsare fittedbyPLSregression. Substantial and reliable lowering of detectable concentration oflanthanidesincomplexmixturescanbeobserved.TheproposedapproachcanbeappliedinsimilarsituationswithXRFmeasurementsinotheranalytesanditsignificantlybroadenspossibleapplicationrangeoftheEDXmethod.

40

P01.MultivariateregressiontechniquesandNMRspectroscopytoquantitativelydeterminefalsificationofpharmaceuticalandfoodproductsYuliaB.Monakhova1,2,BerndW.K.Diehl21InstituteofChemistry,SaratovStateUniversity,Saratov,Russia2SpectralServiceAG,Köln,Germany

Thereisnodoubtthatmultivariateanalysisisrequiredtoauthenticatecom-plexpharmaceutical samples and foodproducts and, especially, toquantita-tivelypredicttheamountofadulterationinblends.Inthisreportweconsidertwoprojects aimedat authenticationofporcineheparinand sunflower leci-thin by NMR spectroscopy. For multivariate modelling several statisticalmethodssuchaspartialleastsquaresregression(PLS),ridgeregression(RR),stepwise regressionwith variable selection (SR), stepwise principal compo-nentregression(SPCR)wereutilizedformodelingNMRdataofin-housepre-paredblends(n>80ineachcase).

In caseofporcineheparinblendedwithbovine speciesPLSandRRshowedthebestperformancewiththelimitofdetection(LOD)androotmeansquareerrorofprediction (RMSEP)below2%w/wand1%w/w, respectively.Re-producibilityexpressedincoefficientsofvariationwasestimatedtobebelow10%. The developedmethodwas found to be applicable also to heparinoidmatrix(notpurifiedheparin).

Themodelsbasedonphospholipid (31PNMR), fatty acid (1HNMR)and sac-charide(1HNMR)distributionswerevalidatedtodistinguishpuresunflowerlecithinfromthatblendedwithsoyspecies.PLSbasedonsaccharidecomposi-tionisabletoestimatelecithinfalsificationregardingitsvegetableoriginwiththesensitivityandRMSEPbelow1%w/wand3.5%w/w,respectively.Pre-dictionerrorwasimprovedbyfusingthreeavailableNMRprofiles.

The developed approaches allowed to evaluate the composition of lecithinandheparinblendsonthebasisofmultiplechemicalcomponents inparallelwithofficialNMRtests.

41

P02.DiagnosticsofsinteringprocessesonthebasisofPCAandtwo-levelneuralnetworkmodelE.G.Egorova,I.V.Rudakova,L.A.RusinovSt.PetersburgStateInstituteofTechnology(TechnicalUniversity)

Themanufacturingprocessofironoresinterflowsundertheinfluenceofthesetofdisturbances,includingchangesinchemicalandmineralogicalcomposi-tionandparticlesizeofthecharge,theconditionsofitsdosingandplacingonthesinteringbelt,variationsinthecharacteristicsofthefuel,ignitionandsin-teringconditions,etc.Controlsystemsstabilizethemainprocessparameters,without considering the frequently occurring deviations from the nominalvaluesofthesefactors.Thisleadstowastageofsinterquality,andthusreduc-estheeffectivenessoftheprocess.

Surveyofexpertsandanalysisofoperatingregulationshasallowedidentify-inganumberofabnormalsituations,eachofwhichiscausedbyvariousfaults.The taskof thediagnostic system is to identify these faults in on-linemodeandgivingouttherecommendationstooperatorsforreturningtheprocessinnormalstate.Thesystemoperatesinparallelwiththetraditionalcontrolsys-tem.

Fromthevarioustypesofdiagnosticmodelsappliedforsolvingtheproblemsof diagnosing complex systems, towhich the sintering process also relates,neuralnetworkmodelwasselected.Theformingofabnormalsituations,eachof that brings together a group of faults with similar symptoms, led to theformation of the hierarchy in the structure of the diagnosticmodel. An im-portantadvantageofsuchahierarchicalneuralnetworkmodelisasignificantreductionindimensionsofnetworksformingthemodel,whichmeansasharpreduction in network learning time and, more importantly, the time of re-learning,thatissometimesnecessarytoadaptthemodeltothechangingcon-ditionsoftheprocess.

Thetop-levelnetworkperformsthetaskoffaultlocalization.Faultidentifica-tion ismadeby low-levelnetworks.To furtherreducingof thedimensionofthetop-levelnetworkthePCAisused.Thetrainingofthetop-levelnetworkiscarriedoutbythedataofthematrixofPCAscores.Thelow-levelnetworks,inprinciple,havesmalldimensions,sotheinitialvariablescanbeusedthereas

42

inputnetworkvariablesthatfacilitatetooperatorstheperceptionofdiagnos-tic results. To prevent a situation when the network loses already definedfaultduetolackofresultsoftheoperatoractions,i.e.whenthesituationcon-tinues to evolve, the changing of process parameters on the inputs of net-worksshouldberestricted.

ThePCAallowsusingtheclassicmonitoringoftheprocess,bycontrollingthestatisticsT2andQ,andstartingthediagnosticsafterthedetectionofthefaultoccurrence,thatincreasessystemreliability.

Experimentalstudyofthesystembysimulationontherealdatashowedeffi-ciencyofproposedmethod.Thesystemconfidentlydefinesfaultswithin20%oftheirmaximumdevelopment.

P03.CombinedanalysisofX-raydiffractionpatternsandIRspectraofLiFePO4AnnaYarchuk1,EvgenyKarpushkin1,AndreyBogomolov21MoscowStateUniversity,Moscow,Russia2ArtPhotonicsGmbH,Berlin,Germany

Lithium ironphosphate (LiFePO4)hasbeen recognizedasapromising cath-odematerialforapplicationinlithium-ionrechargeablebatteries.Thischem-istryprovideslongcyclelife,veryconstantdischargevoltage,andtheenergydensitycomparabletothatofthecommonlyusedLiCoO2-basedcathodes.Ad-ditional advantages of LiFePO4 over other cathode materials include lowercostandimprovedsafety.Ontopofthat,high-qualityLiFePO4intheformofnanosized crystalline particles can be prepared via a simple hydrothermalsynthesismethod.

Importantly,theelectrochemicalpropertiesofLiFePO4arehighlysensitivetoitsparticles size and shapeand thepresenceof impurities, especially at thecrystals surface. These material characteristics depend on the preparationprocedure.Luckily,thehydrothermalsynthesisisfairlyflexibleandallowsforpreparationofLiFePO4particlesofdifferentmicromorphology.

X-raydiffractionanalysishasbeenacceptedasastraightforwardtooltoeval-uate thequalityof LiFePO4product.However, our trialshave indicated thatlithium ironphosphatespecimenspreparedviasignificantlydifferentproto-

43

colsexhibitsimilarXRDpatterns.Therefore,wehaveappliedIRspectroscopyasasupplementarytechniquefortheproductstudy.ThedetailedanalysisoftheIRspectra(includingthebandsassignment)isfairlycomplicated;howev-er,wehavedemonstrated that thevibrational spectraarevery indicativeofthematerialpurityand(likely)morphology. In thiscontributionwepresentthefirstattemptofcombinedanalysisofXRDpatternsandIRspectraofvari-ousformsofLiFePO4assistedbyelectronmicroscopyimagesanalysisinordertodevelopanapproachtowardsfastandreliablecharacterizationofLiFePO4andsimilarmaterials.Inparticular,wehaveperformedclusterandsimilarityanalysis of the IR spectra and X-ray diffraction patterns and demonstratedthatIRspectroscopyisadvantageousoverX-raydiffractionmethodtoevalu-atetheLiFePO4productquality.

P04.Optimizationofinformationsupporttoenhancediagno-sabilityoftechnologicalprocessesS.A.Kosarevskaya,V.V.Kurkina,L.A.RusinovSaint-PetersburgStateInstituteofTechnology(TechnicalUniversity),SaintPe-tersburg,Russia

Alargenumberofsensorsaregenerallyaspiredtousetoensuretheeffectivemonitoringoftheprocess,sinceitconsideredthatthemorethesensors, thebetterqualityofmonitoringisachieved,andthefaultswillbedetectedmoreconfidently.However,economicandtechnicalpossibilitiesimposesignificantlimitationsonthenumberofsensorsrequiredformonitoringanddiagnosisoftheprocess.Firstofall,itisthecostofpurchasingandinstallingthesensors,inadditionthetechnicallimitationscanbe,forexample,duetotheimpossibil-ityofdirectmeasurementofavariable,etc.

Itisimportantnotonlytominimizethenumberofsensors,butalsoeffectivelyplace themon theprocess for increasing theobservabilityofprocess states,detection and identification of its possible faults. Different approaches areusedfordeterminingtheoptimumnumberofsensorsandtheirlocation(cre-atinganetworkofsensors)forensuringearlydetectionandcorrectidentifi-cationofallpotentialprocessfaults.Themostpopularapproachusesdirectedgraphs.Thegraphnodesrepresentthevariablesoftheprocess,andtheedgesshowacausaleffectbetweennodes.

44

Analysisof thegraphstarts from thenode,where it isassumed theappear-anceofafault(therootnode),andthenthesearchforothernodesreachablefromtherootoneiscarriedout.Ifonthiswayonthegraphatleastonenodewiththesensorwillbefound,itcanbeassumedthatthisfaultwillbedetect-ed.Sothesensornetworkwillbeformed,butitwillnotbeoptimal.

Somesensorsof thisnetworkareobligatoryandcannotberemoved, forex-amplesensorsthatareusedforcontroltheprocessparameters;onlytheposi-tionandthenumberofsensorsthatwillbeuseforimprovingtheobservabil-ityoftheprocessandincreasingthedepthofdiagnosticscanbeoptimized.

In this case the classic genetic algorithm is preferable among the stochasticalgorithmsforthesolutionofmultiextremaloptimizationtasks.Thesolution(chromosome)here isanetworkof sensors; thepopulation isa setof chro-mosomes.Goodnessofsolutionsinthepopulationisdefineduponthefitnessfunctionthatiscalculatedastheinversevalueoftheglobalpenaltyfunction.Penalties are assigned in proportion to the cost of the installed sensor andlossesowingtonotdetectingthefault.Afterassessingfitnessofallthechro-mosomesinthepopulation,anewgenerationiscreatedfromthispopulationbyusingtheselection,crossover,andmutationoperators.Thestopoftheal-gorithmisdefinedbytheabsenceofimprovementsinseveralsuccessivepop-ulations. The obtained solution comprises aminimum sufficient sensor net-work,theuseofwhichallowsobtainingtheeffectofearly-detectionoffaultswithoutcostlymeasuring.

P05.Voltammetricidentificationofantiarrhytmicdrugsusingprincipalcomponentanalysis

R.A.Zilberg,Yu.A.Yarkaeva,A.I.Khabletdinova,A.V.Sidel’nikov,V.N.Maistrenko.BashkirStateUniversity,Ufa,Russia

Voltammetricmethodsarewidelyusedforthequalitycontrolofpharmaceu-ticals,becausemanymedicinalsubstancesareelectrochemicallyactiveintheavailablepotentialregion.Inthiswork,possibilitiesofvoltammetryarecon-sidered as applied to problems of the identification of antiarrhytmic drugsfromthegroupofβ-adrenoblockers(atenolol,propranolol,sotalol,metopro-

45

lol,amiodarone)fromdifferentproducersusinganelectronictonguebasedonglassycarbonelectrodesmodifiedwithpolyarylenephthalides(PAFs).

Toassessthesimilaritiesanddistinctionsbetweenthecyclicvoltammograms,weusedtheirPCAmodeling.Inmostcases, inthescoreplotsofPCAmodelsonthecoordinatesPC1–PC2,thevoltammogramsofmedicinesfromdifferentproducers belonged to different clusters, depending on the nature and con-centrations of auxiliary substances. The optimumnumber of principal com-ponentswas selected so that explaineddispersionwasnot lower than90%(2–5 principal components). The stronger was the difference between thecompositionsof auxiliary substances in somemedicines, the longerwas thedistancebetweenthecorrespondingclustersintheplaneofprincipalcompo-nents.ThescoreplotsofPCAmodelingwereusedtogaininformationaboutthe complianceofmedicines to the standard samples and also for revealingthe distinctions between the products from variousmanufacturers. In addi-tion, the voltammograms ofmedicines transformedby principal componentanalysiscanserveas“virtualprints”andbeusedfortheir identificationandqualitycontrol.

Itshouldbenotedthattheidentificationofmedicinesdifferinginthecompo-sitionofauxiliarysubstancesusing three-factor images (registrationofvolt-ammogramsonthreeglassycarbonelectrodesmodifiedbyPAFs)allowedustoconsiderablyincreasethepercentageofcorrectlyrecognizedsamplescom-paredtotheregistrationofvoltammogramsononlyoneelectrode.Inthelat-tercase,clustersofmedicinesfromvariousproducersoverlappedeveninthecasewhenmedicinescontaineddifferentauxiliarysubstances.

ThisworkwassupportedbytheRussianFoundationforBasicResearch,pro-jectsnos.14-03-97067-r_povolzhye_aand15-03-01388-a.

46

P06. Mathematical modeling of "composition-current"phasediagrams for three-componentheterogeneous alloysof cadmium-tin-bismuthMoshchenskayaE.Yu.Samara StateTechnical University,Samara, Russia

The phase diagram is one of the main tools in solving most of the problems of modern physical and chemical analysis in materials science, metallurgy and chemical engineering. Currently, methods of experiment planning are widely used in physical and chemical analysis particularly for charting "structure-property". Experimental design by thermodynamic modeling of invariant points on the chart "structure-property" greatly simplifies the experimental study of multicomponent systems.

The present work is dedicated to the development of methods and algorithms for the simulation of phase complex of physical and chemical systems as well as to automation of the research and analysis process. The method of local electrochemical analysis (LEA) uses "composition-current" phase diagrams as a source of information. Dependencies of the partial currents phase dissolution on the alloy composition are used as calibration curves.

I have offered equations for modeling "composition-current" diagrams for different alloy systems. An algorithm, which calculates the partial currents within the composition of the alloy components, has been developed. This algorithm was implemented in a visual programming environment Delphi.

The equations for computing the current partial dissolution for cadmium, tin and bismuth are obtained for the dissolution process from matrix metal ternary alloy Cd-Sn-Bi. Points for each of the three surfaces were calculated using a program written in a visual programming environment Delphi.

Sets of coordinates (x, y, z) in the form of an array of numerical data for each of the three surfaces were obtained. The program allows to calculate the coordinates with steps specified by a user. The resulting sets of coordinates were visualized using software Algebrus 3.1.

It has been established that the intersection point of three surfaces on the "composition-current" diagram in a ternary system Cd-Sn-Bi corresponds to a relative minimum in the total dissolution of the components surface currents.

47

P07.Recognitionofphysiologicalstatesusingstabilometricda-ta:afeasibilitystudyAndreyA.Muraenko,SergeiI.ZhilinAltaiStateUniversity,Barnaul,Russia

The principle of stabilometry consists in translating the mechanical oscillations of man's physiological gravicentre into electrical signals by means of force plates, signals amplification, recording, and analysis [1]. Traditionally, the stabilometry was considered as a technique for measuring only the balance of human beings un-der certain conditions and diagnosing balance-related disorders. As the stabilome-try is simple and rapid to use, stabilometric data is potentially suitable for research and applications not only in clinical medicine but in broader context as well. The presented work is aimed at the discovering features of stabilometric time series and studying their informativity and usefulness for physiological states detection and physical activity for the development of systems for monitoring and control of physical activity in sport and rehabilitation.

The first contribution is the way to construct features of stabilometric signals suit-able for distinguishing of physiological states. The idea is to collect and aggregate scalar and histogram statistics of signals for running time-window: amplitude, standard deviation, etc). The resulting feature vector is obtained by concatenation of features for several window widths. Discrimination of different object (physio-logical states) in principal components space was studied for the following groups of persons: 1) patients of neurological clinical unit after hypertension stroke sur-vived in 2-5 days before stabilometric measurements; 2) adult patients with infan-tile cerebral paralysis which are able to stay at force plate without assistance; 3) adults in good health (without affirmed health problems). The results of the study showed principal feasibility of the proposed features of stabilometric time series for pairwise distinguishing of mentioned groups of person.

The second contribution concerns the feasibility of stabilometric data statistics for estimation of sportsman organism response to exercise stress. Stabilometric meas-urements were collected during power training of sportsmen with different level of preparation before and after each exercise. It is concluded that the mean density of stabilometric trajectory is correlated to the change of physical activity and can be substantially interpreted by experts.

48

References Chiari L. Stabilometry. In: Binder M., Hirokawa N. and Windhorst U. Encyclope-dia of Neuroscience. Berlin Heidelberg: Springer, 2009. pp. 3830-3833.

P08.Identificationofbisoprololfromdifferentmanufacturersusingvoltammetric“electronictongue”Yu.A.Yarkaeva,R.A.Zilberg,E.I.Maxyutova,A.V.Sidelnikov,andV.N.MaistrenkoBashkirStateUniversity,Ufa,Russia

Due to the increased requirements for the quality of medicines, the im-portanceoftheuseofmodernstandardizedmethodsfortheiranalysisgrows.Alongwith the use of chromatographic, opticalmethods of determining or-ganic compounds electrochemical analysis methods are commonly used, inparticularvoltammetry.

Inrecentyears,theformationandpracticalapplicationofchemicallymodifiedelectrodesbecameinteresting.Sensorsbasedontheseelectrodesareusedfordeterminationofalargenumberofdifferentsubstances,particularlymedici-nal compounds.Formodifyingdifferentchemical compounds,polymer filmsareappliedtothesurfaceoftheelectrodematerial.Theysignificantlychangetheabilityoftheelectrodetothevoltammetricresponse.Suchmodifiedelec-trodesareoftenusedinmultisensorysystemsas"electronictongue".

New three-electrode multisensory system are proposed for voltammetricanalysisofantiarrhythmicdrugsusingthreepolyarylenephthalidesasmodifi-ers (TO, TOO, TOOO). The use of three different modifiers provides cross-sensitivityconditionofelectrodesrequiredforthefunctioningofthevoltam-metricsystemsasthe"electronictongue".

InthisworkoriginaldrugofbisoprololKonkoranditsgenerics(Aritel,Biprol,Niperten,Bisogamma,Bisoprolol,Bidop,Biol,KonkorKor,Kardinorm)wereinvestigatedbyvoltammetricmethod.Processingof theobtainedvoltammo-gramsofantiarrhythmicdrugswascarriedoutusingchemometricmethods:principalcomponentanalysis(PCA)andmethodofsoftindependentmodelingofclassanalogy(SIMCA).

49

Theuseofpolyarylenephthalidesasmodifiersprovidesaclearseparationofthe response signals of antiarrhythmic drugs on the plot of PCA-modeling.SIMCA-classification showed that theproportionof correctly identified anti-arrhythmic drugs varies between 67-100%. Thus, express-identification ofmedicines using a voltammetric "electronic tongue" canbe carried outwithhighaccuracy.

ThisworkwassupportedbytheRussianFoundationforBasicResearch,pro-jectnos.15-03-01388-a.

P09.SoftwareimplementationoftheData-DrivenSIMCAmethodY.V.Zontova,O.Ye.Rodionovab,A.L.PomerantsevbaNationalResearchUniversityHigherSchoolofEconomics(HSE),Moscow,Rus-siabSemenovInstituteofChemicalPhysicsRAS,Moscow,Russia

We present a software implementation of the Data-Driven SIMCA method(DD-SIMCA)[1,2].Themethod is implemented inMatlabscripting languageintheDD-SIMCAToolbox,whichprovidesthemeansfordatapre-processing,interpretation, validation and visualization. The two main modules of thetoolboxarenamedinaccordancewiththeirbasicfunctionality,DD_SIMCAandDDS_PLOTrespectively.Thefirstoneisusedforthetrainingandtesting.Thesecondoneisemployedforgraphicalrepresentationandanalysis.Bothmod-ulesrelyonthefixedstructuredrepresentationoftheclassificationmodelaswellasthevalidationresults.Thisapproachsimplifiesthecommunicationbe-tweenthesoftwaremodulesandthereusabilityofthemodel.

Themainfeaturesofthesoftwareareasfollows:

• The DD-SIMCA method is implemented using the best Matlab program-ming practices. The command-line interface provides numerous options,whicharewelldocumentedandexemplified.

• The Acceptance plot provides a graphical representation of the decisionarea,normalsamples,extremesandoutliersinauser-friendlymanner.TheExtremeplotshowstheobservedvs.theexpectednumberofextremeob-jectsandprovidesgraphicalmeansforthedataanalysis.

50

• Calculationoftherateofthefalseacceptedsamples(typeIIerrorβ)forthenewclasses.

• ThetoolboxmaybeusewithouttheMatlabStatisticsToolbox.

Thesoftwarefunctionalityisillustratedwithareal-worldexampleofquanti-tativeriskassessmentinclassificationofdrugswithidenticalAPIcontent[3].

References

1.A.L.Pomerantsev,O.Ye.Rodionova,J.Chemometrics,28,429–438(2014)

2.A.L.Pomerantsev,O.Ye.Rodionova,J.Chemometrics,28,518-522(2014)

3. O.Ye.Rodionova,K.S. Balyklova,A.V. Titova,A.L. Pomerantsev, J. Pharm.Biomed.Anal.98,186-192(2014)

P10.TheuseofPLS-DAandpulsedelectricstimulationmethodforassessingthequalityofbiologicalobjectsAlekseyG.Zarubina,b,YuliyaN.Dolgovab,c,AnatoliyI.Mamaevb,caDepartmentofOilandGasStorageandTransportation,TomskPolytechnicUniversity,RussiabDepartmentofAnalyticalChemistry,TomskStateUniversity,RussiacSibsparkLimitedLiabilityCompany,Tomsk,Russia

Foodproductshavetoconformtomodernrequirementsforqualityandsafetyof food stuffs. In this regard the development of identificationmethods forfreshbiologicalproductionandbiologicalproductionafter freezingandsub-sequentthawingrepresentsacademicinterest.

Asbiologicalproductionthesamplesofafreshcrucianandanunfrozencru-cianwereselected(thefrozenmeatofthecrucianafterfreezingwasdefrost-ed in the thicknessofmuscular tissueupto the temperaturenot lowerthanminus1 °C).To receiveananalytical signal themethodofpulseelectric im-pact on biological fabric withmeasurement of electric current force valuesdependingonthetimeperiodofelectric impactwaschosen[1].ThePLS-DAmethodwaschosenforaformalclassification.Selectionofexperimentaldatawasexpandedwiththemethodofnumericalstatisticalmodeling(theMonte-Carlomethod)onthebasisofaprioristicstatisticalinformationabouttheana-lytical signal. On the basis of the obtained model data with the use of

51

CHEMOMETRICS library [2] the PLS-DAmodel for two classes of biologicalproduction – the fresh crucian and the crucian after freezing and thawing–was constructed and checked. Itwas established that thePLS-DAmodel fortwohiddenvariableswascharacterizedbythe lackoferrorsof thefirstandsecondkindforpredictionofclassesforthefreshandunfrozencrucian.

As a result of the conducted research itwas established that themethodofpulseelectricimpactonbiologicalfabricwithmeasurementofelectriccurrentforcevaluesdependingon the timeperiodofelectric impact incombinationwith the PLS-DAmethod allows carrying out unambiguous identification ofthefreshandunfrozencrucian.

References

1.Dolgova,Yu.N.,A.K.Chubenko,andA. I.Mamaev.Pulsedelectricstimula-tionasamethodforassessingthequalityofbiologicalobjects.RussianJournalofGeneralChemistry84.5(2014):1061-1064.

2.PomerantsevA.L.,ChemometricsinExcel,JohnWiley&Sons,2014.

P11.WeightselectionandrotationambiguityinmultivariatecurveresolutionofabsorptionspectrawithweightedleastsquaresSkvortsovA.N.SaintPetersburgPolytechnicUniversity,St.Petersburg,Russia

Multivariatecurveresolution(MCR)issetofchemometrictechniques,whichaim to resolve data matrices into spectra and concentrations of individualchemicalcomponents,givenasetofphysicochemicalconstraints.GivenadatamatrixXanMCRmethodsearchesforabestbilinearmodelCST,whereCandSareconstrainedmatricesofspectraandconcentrations.InmostMCRmethodsthisisdonebyminimizingthesumofsquares∑mne2mnoftheelementsoftheresidualmatrixE=X–CST.Suchapproachimpliesthattheresidualsaredistrib-utednormally,similarly,andindependently.Thisisnotalwaysthecaseinrealdata,where theuncertaintiesmaydiffer frompoint topointand/oroutliersmayoccur.Oneoftheclassicsolutionsistouseweightedleastsquares(WLS)approach,whichminimizes∑mne2mnwmn,where{wmn}isaweightmatrix.Typi-cally, theweightmatrix is constructed from themaximum likelihoodprinci-

52

ple.However,WLS isavery flexible technique, itcanbeadapted forMCRofsparcematrices,robustMCR(byiterativereweightingprocedure),andsomenonlinearMCRtasks.

AcomplicacyofMCRwithWLSistheinapplicabilityofprincipalcomponentsfor finding MCR factors, so many good MCR methods cannot be used withWLS. The problem can be solved by weighted alternative least squares(termedMCR-WALS).Also a fasterpositivematrix factorization (PMF) algo-rithmwasproposed(Paatero,Tapper,1994),whichiscurrentlyawidelyusedWLSmethod.ThepossibilitiesofMCR-WALS,PMFandsomeothermethodswererecentlyrevisitedin(Stanimirova,Tauler,Walczak;2011).MCR-WLSso-lutions,similarlytocommonMCR,mayhaveresidualrotationambiguity(RA).WhileRA in commonMCRwas extensively studied,RA inMCR-WLSand itsrelationtoweightmatrixismuchlessknown.

InthepresentworkwefocusedonMCRof‘saturated’absorptionspectra(op-tical density). Optical densityD,which is bilinearwith rellllspect to spectraand concentration, is a log-ratiomeasurement. Togetherwith the nature ofuncertainties in light intensitymeasurements this results in strong depend-enceofmeasurementuncertaintyδDonDwithtypicalsharpincreaseofδDatlargeD.SolargeDvaluesareunstable,effectivelynonlinear(hencetheterm‘saturatedspectra’)andcompromisetheMCRmodel.CommonMCRisunabletohandlethissituation,butMCR-WLSmaysucceedifproperweightsarese-lected,whichwoulddamporevenzerooutthecontributionoflargeDvalues.Still,theanalysisofresidualRAofthesolutionisrequired.

TotesttherelativeefficiencyofdifferentweightingschemesandestimateRAinMCR-WLSseveralsimulatedexamplesandareal-lifedatamatrixwereana-lyzed.MCR-WLS factorsandanalogsofprincipal components (forRAanaly-sis)were calculatedbyMCR-WALSandweightedversionofNIPALS respec-tively.RAwasestimatedbyBorgenplots,tracking,andchargedparticleopti-mization cPSO. Simple threshold weighting (w=0 | D > Dmax) or accurateweighting based on uncertainty function andmaximum likelihood principlewereused.

Bothweighting schemes displayed better performance inMCR of saturatedspectrathanunweightedMCR(regardingthedeterminationofthenumberof

53

components, retrieving spectra and concentrations, and satisfying the con-straints).MostoftheexistingmethodsofRAestimationcanbegeneralizedtoMCR-WLS. RA in simple MCR and MCR-WLS for slightly saturated spectrashowedsimilarbehavior,althoughtheshapesof the feasibleareasweredif-ferent.Strongsaturationinthespectraincreasedrotationambiguity;itcorre-spondedwell to thedecrease in therelativecountof ‘reliable’and ‘extreme’data points. It was also shown that accurate weighting of measurements,whichareclosetozero,canexpandthepowerofBorgenplotRAanalysis(asanalternativetogeneralizedBorgenplotalgorithm).

P12.Themethodofhyperspectralimageclassificationforiden-tifyingoilpollutiononthegroundM.Kuznetsov1,E.Ramenskaya2,V.Ermakov21MoscowInstituteofPhysicsandTechnology,Moscow,Russia2SamaraStateTechnicalUniversity,Samara,Russia

Theproblemofenvironmentalpollutionwithoilandoilproductsisnotnew.Questions of rapiddetection and liquidationof oil spillages remain relevantformanyyears. It isproposedthat theapproachofusingremotesensing, inparticular,theanalysisofhyperspectralimageoftheEarth'ssurfacewillallowtosolvetheproblemofenvironmentalmonitoringofoilpollution.Somediffi-cultiesariseintheanalysisofhyperspectralimages.Weaddresstheproblemoftheoilspillidentificationonthelandsurfaceusingthehyperspectralimag-esprocessingmethodology.

Theconsidereddataarethehyperspectralimagesofthelandsurfaceobtainedfromtheairbornelab,withtheexpert-givenallocationoftheregionsofinter-est.Thatis,theexpertsallocatetheoilspillareastogetherwiththeareasthatdon’tcontainanyoilspillswithcertainty.Thesecond-typeallocationsincludethe dark regions similar to the spills: infrastructure shadows, small waterponds,cloudshadowsetc.

Usingthegivendata,weconstructatwo-classclassificationmodeloftheim-agespectra.Thefirstclasscontainstheallocatedoilspillspectra,thesecondclassconsistsofspectraoftheremainingobjects.

54

The developed classificationmethod consists of the two stages. On the firststageweestimateaGaussianmixturemodeldescribingthegivenspectralda-ta.Usingthemixtureweperformdataclusteringandcomputetheclustercen-ters.Onthesecondstageweconstruct theclassificationalgorithmusing thedistancestotheclustercentersasthenewfeaturedescriptionofthespectra.

Theproposedmethodallowstoachievehighclassificationaccuracy throughtheuseof theclusterdataassumption.Theconstructedmodel issimpleandinterpretable:theoilspillscorrespondtothespectrawiththesimilarshapetothe cluster centers of the allocated spills. Furthermore, the training stage iscomputationallyefficient: it is linearoverthetotalobjects(theallocatedhy-perpixels) number. This computational requirement is essential for the hy-perspectraldataprocessing.

Ourcurrentandfurtherinvestigationsaredevotedtothecollectingandpro-cessingthesatellitehyperspectralimages.Uponachievingacceptablerecogni-tionquality,wewillproviderecommendationsonmodificationsoftheecolog-icalmonitoringsystemsinoilandgascompanies.

P13.RamanspectrometryforpharmaceuticalwatersolutionsanalysisMaximKhannanov1,2,,OlegVolkov2,RomanGaleev3,LeonidKulik1,2,NailKhan-nanov2,RamilIbragimov2,SergeyDorofeenko2.1InstituteofSolidStatePhysicsRussianAcademyofSciences,Chernogolovka,Russia2EnSpectrLCC,Chernogolovka,Russia3Federalserviceofhealthcare,Moscow,Russia

Certification, counterfeit revealing, raw material inspection and so on aretasks related to drug solutions identification. The problem in this case is toperform it routinelyandduring short timeasdefinedby thecontrollingau-thorities.

Chromato-mass-spectrometrydoesnotallowpracticallytoperformtheiden-tification routinely for the complexity of themethod itself, lack of portablespectrometermodification, requiredsamplepreparationaswell as theneedofahighlyqualifiedoperator.Chromato-mass-spectrometryanalysisrequires

55

aswell ampoules opening that is generally forbiddenwithout themanufac-ture’spermission.

Ramanspectrometrydoesnothavethesedrawbacksbeingarapidlyexpand-ing non-destructive method for solid and liquid samples analysis. Ramanspectroscopyallowsmeasurementsofliquidandpowdersubstancesthroughatransparentorsemi-transparentpackagewithoutitsdestruction,theidenti-ficationofpuresubstancesandmixesinamatterofseconds.Themainprob-lemthoughaboutquantitativeRamananalysisofwatersolutionswithporta-blesetupisthedetectionofwateritself.

WereportonexperimentalresultsofRamanspectraanalysisofpharmaceuti-calwatersolutionswiththeconcentrationofaboutonepercentwithunprec-edented accuracy in an intact ampoule using a specially developedportableRamanspectrometer.TheinvestigationofwaterRamanbanddependingonapH-value, the luminescence andRaman scattering froma glasspackage, theanalytespectrachemicalshiftinwaterareperformedinordertoarchivethementionedcharacteristics.

Inconclusionanewtechniqueisgivenforthedistinguishingbetweendiffer-entmedicines inwater solutionsand followingcalculationof theconcentra-tionofthemainandadditionalcomponents.ThistechniquewassuccessfullytestedbyRussianfederalserviceofsurveillanceinhealthcareandprovedallthementionedcharacteristics.

P14.ApplicationofChemometricsinGeoecologyK.Chertes,O.Tupitcina,V.ErmakovSamaraStateTechnicalUniversity,Samara,Russia

SamaraStateTechnicalUniversityhas a great experience in solvinggeoeco-logicalproblems.Geoecologyisafundamentalandappliedsciencethatdealswith theengineeringobjectsandgeologicenvironment cooperation.Bothoftheelementsgeneratethedestructedgeologicalsystems(DGS).Forexample,off-site landfills,borrowpits,contaminatedandpolluted lands,uncompletedconstructionprojectsandtheiragglomerations.

TheimportantgeoecologicalproblemistheDGSrecovery.Sinceitisimpossi-bletorecoversomeoftheDGS,agreatnumberofparametersshouldbeesti-

56

matedbeforehand.Thoughit’spracticallyimpossibletoidentifythedegreeofinfluenceoftheparameters.

DGSare themultivariatedatasystems.So thedominantcomponentanalysis(DCA) is themost useful instrument for their estimation. ThismethodwasusedfortheestimationofmorethanhundredsofDGS:open-casts,wastede-posits,sludgetanks,sludgedrainingbedsandotherartificialgeo-structuresinthewater industry,waste handling services, agriculture, industrial and civilconstruction.

Chemometricmethods are constantly used as themathematic and practicalfoundationof thePhDthesisofSSTUscientists,suchas IRspectroscopyandPCA applied for polymer sorting in thedomesticwaste, application of scan-ningprobespectroscopyinthemonitoringofoilpollution,ecologicalassess-mentoflandfillswithmultivariateanalysis,principlesofmanagingecological-lysafearchitecturalreconstruction,etc.

Astheresultoftheconductedresearchweconcludedthatthemainproblemsofmultivariatedataanalysisusageingeoecologyarethefollowing:

1.Thedifficultyofobtainingrawdata.Thetestobjectsareofbigsize,anditisnecessarytocarryagreatnumberoflaboratorywork.

2.Theinabilitytouseordinaryanalyzers.Thestudiedobjectsaremulticom-ponentandmultiphase.

3.Thestudiedsystemsareopen.Theobjectsareinextricablylinkedtoallthecomponentsoftheenvironment.

4.Thedynamicsof thestudyobjectdoesnotdependontheresearcher.Therateofchangeinthepropertiesofairandwaterishigh.Thegeologicalenvi-ronment is changing slowly. The abrupt changes (rainfall, volley emission,etc.)arepossible.

Notwithstandingtheabove-mentionedproblemsweshouldsaythatthestudyofcomplexgeoecologicalsystemswithouttheuseofmultivariatedataanaly-sisisimpossible.

57

P15.EnvironmentalpollutionassessmentusingPCA/PLSanal-ysis(throughtheexampleofsoilandnaturalwaterofKaraidel'districtofBashkortostanrepublic)KabirovaL.R.1,SidelnikovA.V.2,LyginS.A.11BirskbranchofBashkirStateUniversity,Birsk,Russia2BashkirStateUniversity,Ufa,Russia

Biotesting isadeterminationof toxicitywith thehelpof testers.The testersgivesignalsofdangerregardlessofwhatsubstancesandinwhatcombinationcausechanges invitalsignsof the investigatedorganismsand, therefore, forrevealingthemainfactorsofcontaminationfurtherinvestigationisnecessary.

Inthepresentworkweanalyzedtheinfluenceofsoilcompositionandnaturalwater on the indicators of environmental pollution (usingwater from riverKaraidel’andameltwater).PrincipalComponentAnalysis(PCA)wasappliedtoestablishchemicalfactorsofsoilheavymetalspollutionanddeterminethebehaviorofsuchtestorganismsasbuttercup(leadpollution)andwatercress(copper,zinc,cobaltpollution)growingattheAkbulyak-Baiki-Birskhighway.

Partialleastsquares(PLS)discriminantanalysiswasemployedtoinvestigatethe influenceof certain indicators (e.g., ionsofheavymetals, COD,BODandammoniumgroups)onthelevelofpollutioninthesamplingareas:villagesofBerdyash and Horoshaevo located near the river Karaidel'. K-criterion wasusedas a generalized criterionof thenatural objects contamination level. Itwasequatedto1forobjectswithmaximumpermissibleconcentration(MPC)contaminationlevel,andto0forobjectswithcontaminationlevellowerthanMPC.ThePLSmodelwasbuiltusingtheK-criterionasaresponsevariable.ItwasfountoutthattheconcentrationofharmfulsubstancesintheareasunderstudydoesnotexceedtheMPC(thepredictedKvaluesdidnotexceed0.2).

58

P16.Calibrationtransferbetweendifferenttypesofmultisen-sorsystemsM.M.Khaydukova1,2,D.O.Kirsanov1,2,V.V.Panchuk1,C.Medina-Plaza3,M.L.Rodriguez-Mendez3,A.V.Legin1,21Laboratoryofchemicalsensors,InstituteofChemistry,St.PetersburgStateUniversity,St.Petersburg,Russia2ITMOUniversity,St.Petersburg,Russia3SensorGroup,Departmentofinorganicchemistry,E.T.S.IndustrialEngineering,UniversityofValladolid,Valladolid,Spain

Multisensorsystems,alsoknownas“electronictongues”,arepromisingana-lyticaltoolsforexpressanalysisofvariouscomplexparametersinliquids,likee.g. taste, toxicity, etc. These systems are a combination of hardware andsoftwareparts.Hardwarepartconsistsofchemicalsensorarrayanddevicesforsensorresponseregistration.Twomostcommonlyusedsensortypesarepotentiometric and voltammetric. Software part is related to chemometrictechniques.Typicallymultivariateregressionmodellingisemployedtorelatetheresponseof sensorsystemwithparameterof interest.Agoodmodel re-quires considerable efforts and large number of samples, which can beunique,expensiveandhardtoget.Itwouldbeverypracticaltohaveamethodcapableofmultivariatecalibrationtransferbetweenvariousmultisensorsys-temsworkingondifferentprinciples.

InourpreviousresearchasimplealgorithmbasedonPCAdecompositionhasbeensuggestedforcalibrationtransferbetweendifferentanalyticalmethods.It can convert thedata fromone instrumental format to another [1]. In thisstudyweapplytheproposedstrategytotransfercalibrationmodelsbetweenvoltammetric andpotentiometricmultisensor systems. Sample set consistedof8typesofgrapemusts.Allthesamplesweremeasuredatleastthreetimeswithtwosensorarrays:voltammetricof4sensorswithmetaloxidemodifiedsurfaceoperatedincyclicvoltammetrymode,andpotentiometricof26plasti-cizedpolymericmembranesensors.Cyclicvoltammogramswerecompressedby kernelmethod, 10 kernels for each sensorwere used. Resulting voltam-metric(VA)datawere8×40matrix,whilepotentiometriconewas8×26.TheVAdatawereusedtobuildPLSregressionmodelstopredicttotalacidity,po-tassium concentration, tartaric acid concentration and other parameters of

59

grapemust.RMSEPvaluesofthesemodelsfortotalaciditywerearound0.08(for the range0.66-0.96 lg C(TA)). Potentiometric datawere converted intovoltammetric responseshapeswithsuggestedalgorithmandwereused fur-thertopredictthesameparameters.CorrespondingRMSEPforTAwas0.05(this decrease can be attributed to the higher sensitivity of potentiometricsensors towardsTAwhich isconfirmedbyRMSEPvaluesof theoriginalpo-tentiometricmodelequal0.02).Similarresultswereobtainedforseveraloth-erparameters.

These results show that data conversion between different sensor arraysbased on different working principles is possible and this fact can be em-ployedforbuildingcross-platformmultivariateregressionmodels.

The reported studywas partially supported by scholarship program St. Pe-tersburgUniversityandbankSantander.

References

1. V.V. Panchuk,D.O.Kirsanov,M.M.Khaydukova,V.G. Semenov,A.V. Legin.Latentvariablebasedcalibrationtransferbetweendifferentanalyticalmeth-ods//BookofabstractsCAC-2015,XVChemometricsinAnalyticalChemistry,June22-26,2015Changsha,China–p.25

P17.Methodofimpedancespectroscopyfortitrationofsurfac-tantsusingprincipalcomponentanalysisD.I.Dubrovski,A.V.Sidelnikov,F.Kh.Kudasheva,andV.N.MaistrenkoBashkirStateUniversity,Ufa,Russia

Recently,interestincontrollingthecontentofsurfactantintheoilincreased.In this regard, there is theneed for improvedmethods fordeterminationofsurfactantsincomplexorganicmixtures-oilandwater-oilemulsions.Amongthemspectrophotometryandpotentiometryareusedmostwidely.However,these methods require prior separation, purification, preparation of two-phasesystems.Thisoftencomplicatestheanalysis.

Inthispaper,electrochemicalsystemsandchemometricapproachesthatex-tract information about the content of surfactants in aqueous solutions andwater-oil emulsions using impedance spectroscopy were proposed. An im-portantadvantageof themethod is thehighsensitivityof themeasurement,

60

theabsenceof requirements to theselectivityof theelectrodes, to thepres-enceofcoloredcomponentsandheterogeneousphasesandothers.Thespec-traoftherealandimaginarycomponentsofimpedancewererecordedduringthe titration. The obtained data were processed using principal componentanalysis. Spectrawere converted intopointson aplaneof principal compo-nents.Itisshownthattheinflectionpointcorrespondstotheendpointofti-trationandcharacterizesthestateofequivalencereaction.

ThisworkwassupportedbytheRussianFoundationforBasicResearch,pro-jectsnos.14-03-97067-r_povolzhye_aand15-03-01388-a.

P18.MultivariatediscriminationanalysisofmultispectralfibersensordataforkidneycancerdiagnosticsValeriaBelikova1,UrszulaZabarylo2,ViacheslavArtyushenko2,FranziskaSchulte2,IskanderUsenov3,VladislavGalyanin1,AndreyBogomolov1,2,41SamaraStateTechnicalUniversity,Samara,Russia2artphotonicsGmbH,Berlin,Germany3TUBerlin,InstituteforOptics&AtomicPhysics,Berlin,Germany4GlobalModelling,Aalen,Germany

Accordingtothestatistics,cancerremainsoneofthemostcommoncausesofdeaththroughouttheworld.Surgicalinterventionisstillthemainmethodofmedicaltreatmentofthedisease.Operationsuccessstronglydependsonthedoctor’sability todistinguish tumor fromhealthy tissues.Visualassessmentof tissue sections and biopsy are the commonmethods of histopathologicalanalysis.Itsmaindisadvantagesarethehighcostsandconsiderabletimethatisneededtoobtainresults.Duringasurgicaltreatmentthisputsthepatientatgreaterriskduetoprolongedexposuretogeneralanesthesia.

Overthepasttwodecades,theuseofopticalspectroscopyforbiomedicalap-plicationshasgrownsignificantly.Itsattractivenesscomesfromitsabilitytoprovide quantitative information about the biochemical and morphologicalstatesofthetissueinaminimallyinvasiveornoninvasivemanner.Inaccord-ance with the literature, different fiber-optic probes and spectroscopictechniquesareusedintissueanalysisofskin,breast,lungs,brain,gastricandsomeothertypesoftumor.

61

In thisworkwehave investigatedex- and in-vivo samples ofmalignant andhealthykidneytissuesofelevenpatientsusingfourspectroscopictechniquessimultaneously: mid-Infrared, near infrared, fluorescence and Raman spec-troscopy. The measurements were performed with various fiber-basedprobestransmittingthelightintherespectivespectralrange.Herewereportonthedevelopmentofdiagnosticmultivariatealgorithmscapableofcancerofthehumankidneyusingthismultispectraldata.

P19.SimulationofspectraldataforregressionmethodsvalidationVadimBelikov1,VladislavGalyanin1,AnastasiiaMelenteva1,AndreyBo-gomolov1,21SamaraStateTechnicalUniversity,Samara,Russia2artphotonicsGmbH,Berlin,Germany

Chemometricianstendtousemanytechnicaltrickstoobtainanoptimalmul-tivariatemodel:variousregressionorclassificationalgorithms,dataprepro-cessingtechniques,variable/sampleselectionetc.Thedevelopmentofanop-timaldataanalysisalgorithmisimportantforthefuturemodelperformance,inparticular, in thespectroscopicapplications. Inorder tomake it sure thattheselectedalgorithmsanddatatreatmentsareefficient,theyshouldbevali-dated using spectral data of a necessary nature. The algorithm validationshouldtakeallrelevantspectralcharacteristicssuchasnoiselevelandback-ground, intoaccountandbebasedonasufficiently largeandrepresentativesample set. Often, these requirements can hardly be met using the experi-mentaldataonly.Insuchcases,thealgorithmcanbeverifiedonasimulateddata that canbegenerated ina sufficientvolumeandwith the specifiedpa-rameters. During the algorithm verification itmay be important to find outandtounderstandindividualinfluencesofdifferentfactorsrelatedtothedatanature, suchaspeakoverlap (selectivity), signal-to-noise ratio, thepresenceofanon-linearityandthebaseline,onthefinalresultofanalysis.

Tofacilitatethetaskofalgorithmvalidationwehavedevelopedanapplicationthatgeneratesspectraofindividualcomponentsandtheirmixtures(basedontheBeer’s law)withpreliminaryspecifiedproperties.Thespectraldatagen-erator enables the adjustment of the following parameters: the number of

62

samplesandvariables,thenumberofpeaks,theirshapesandpositions(andhence,thedegreeoftheiroverlap),thenatureandthelevelofnoise,theback-ground intensivenessandnature (multiple scattering, fluorescence, chroma-tographicbaselinedrift)etc.

Here,wepresentthespectralgeneratorsoftwareandshowseveralmodellingexamplesusing simulateddata sets illustrating its advantages for themulti-variate algorithm development and for the optimization of data treatmentroutines.

P20.Comparisonofunivariateandmultivariatecalibrationforleaddeterminationinsoilsbylaser-inducedbreakdownspec-troscopyIvanN.Krylov,SergeyM.Zaytsev,TimurA.Labutin,AndreyM.PopovandNikitaB.ZorovLomonosovMoscowStateUniversity,DepartmentofChemistry,Moscow,Russia,

Laser-induced breakdown spectroscopy (LIBS) is a type of atomic emissionspectroscopywhich employs laser emission for sampling via ablationwhiletheresultinglaser-inducedplasmaactsasbothexcitationsourceandanalyti-cal signal source. Ecological monitoring of lead in soils (for example, nearroads ormetallurgical/glass factories) and geochemical prospecting requirefastelementalanalysisofsolidssuitableforbeingconductedinthefield.LIBSisapromisingtechniqueforsuchpurposes.

However,spectralnoisescausedby interferingelementsworsenaccuracyofoptical atomic emission spectroscopy techniques. Sincemultivariate calibra-tionallowsimprovingaccuracyofanalysiswhensignalsoverlap,wecheckedwhetherthereductionofuncertaintywaspossibleforlowleadconcentrations(32ppmisMPCaccordingtoRussiannorms/16ppmisthemeancontentinEarthcrust)insoilsofdifferenttypes(black,red,graysoilandothers).

A set of 15 sampleswas involved into constructionof calibrationmodels.Aspectralrangecontainingleadlineat405.78nmwasselected.Leave-one-outcross-validationwasemployed tovalidate themodel.Differentbaselinecor-rection methods were tried; subtraction of minimal value within spectralrangeforeachspectrumprovidedthebestaccuracyforPCR.Acomparisonof

63

PCRresultswithunivariateregressionmodelresultswasperformed.TheuseofPCRhasloweredtheRMSEC,thusthedeterminationofleadinsoilsdownto15ppmbecameavailable.Theuncertaintiesofpredictedleadcontentattherequired concentration range were in agreement with the confidence bandgivenforcertifiedvalues.

WethankRSF(14-13-01386)forsupportingtheanalyticalmeasurements;wethankRFBR(15-33-21112-mol_a_ved)forsupportingthepreparationofsam-ples.

P21.Geoecologicalevaluationofthedevelopmentofhousingandcommunalservicessludgestoragesbyhandlingmultidi-mensionaldataV.N.Pystin,N.A.Safonova,V.I.ZuberSamaraStateTechnicalUniversity,Samara,Russia

Activitiesofhousingandcommunalservicesisaccompaniedbytheformationoflow-hazardwaste.Directionsofprocessing,recoveryandrecyclingofwasteofhousingandcommunalservicesarewell-known.Theyaremainlyassociat-ed with the production of building materials and secondary reagents forwastewatertreatment.However,highenergycostsandmaterialcostshindertheir implementation. As a result, thewaste of housing and communal ser-vicesaredepositedinastorage.

Thebasic characteristics of housing and communal services sludge storageswere considered from theperspectiveof their targetdevelopment.The sys-temof conditionappraisalof sludgestorageswas representedby themulti-dimensionaldataanalysis.Thewastedisposalfacilitiesthatwereconsideredinthisstudywereclassifiedintothreegroupsfromtheperspectiveofthefea-sibilityandthepracticabilityoftheirabandoning.Theex-ampleofthedigitalmatrix of thewastedisposal facilities conditionwas given. The constructiveand technologicaldesignof the systemof recultivatingmaterialsproductionwasgiven.

Using composites on the basis of waste of housing and communal serviceswasappliedinthepreparationofareasdisturbedbyobjectdismantlingofin-activebuildingsandenterprises installations in theSamararegion.Theeco-

64

nomic effect of the developed technologies is achieved by reducing the re-quiredvolumesofnaturalsoilandreturnoftherecoveredterritoriesineco-nomicuse.

P22.Attenuatedtotalreflectioninfraredspectroscopyandmul-tivariateanalysis:atoolfordeterminationofoilcontaminationinsoilsAlenaGuryanovaa,VasiliyErmakova,DmitryBykova,ViacheslavArtyushenkob,AndreyBogomolova,baSamaraStateTechnicalUniversity,Samara,RussiabartphotonicsGmbH,Berlin,Germany

Current laboratorymethods for thedeterminationofoil content incontami-natedsoilsprovidenecessaryaccuracyofanalysis,buttheyarenoteffectiveenough for the purposes of environmentalmonitoring. Analysis of the pub-lished data revealed that infrared spectroscopy techniques generally satisfythe accuracy requirements. Being implemented in portable spectrometerstheycanalsoprovidehighspeedandfrequencyofanalysisinthefield.

Herewe report on a study,where themethodof attenuated total reflectionmidinfrared(ATRMIR)spectroscopyhasbeenusedtoquantifythedegreeofoil-contaminationinthesoil.Partial leastsquares(PLS)regressionwasusedas amultivariate calibration tool. The analysis and evaluation of ATRMIR-spectraofmore thanonehundred samplesof the real andmodel soilswithdifferentcontentsofoilandwaterhasbeenperformed.Toimprovethecali-brationaccuracyPLSmodelswereseparatelyconstructedforthesoilsampleswithlow,mediumandhighlevelsoftheoilpollution.Ithasbeenprovedthatsuggestedmethodmeetstheregulatoryrequirementsandcanbeusedfortheinitialdeterminationofoilcontaminationlevelinthesoil.

65

P23.GeoecologicalestimationofsolidwastelandfillswiththeuseofmultivariatedataanalysismethodsSavelievA.A,RamenskayaE.V.MartynenkoE.G.SamaraStateTechnicalUniversity,Samara,Russia

Thenumerousmethodsoflanddevelopmentoflandfillsincluderecreational,forestry-based,landscape,etc.However,thecreationofforestplantationsandparksonthelandfillsterritoriesisproblematicbecauseoftheecologicalspe-cificcharacteroflandfills.Asanalternativetothewidespreadmethodsofde-velopment,theconstructionandeconomicdirectionofterritorydevelopmentcanbe suggested, that is the constructionofwaste sorting stations, areasofballast and recycled fractions briquetting, enterprises on the constructionwasterecyclingintothesecondarygravel,sitesofstackingorgratecontami-natedsoildecontaminationinthesurfaceofwastedisposalfacilities.

Theexistingapproachestotheassessmentofthesolidmunicipalwastecondi-tionusually take intoaccount theirgeometry,humidity,organicmattercon-tent,methanecontent,butthepotentialabilityofarraysolidmunicipalwasteto serve as a construction base of the lowered stability and responsibilitybuildingsisnotconsidered.

Theauthorssuggest the fundamentalprinciples thatprove thepossibilityofconstructionandeconomicdevelopmentofderelictterritoriesanddefinethemeasure groups necessary for the full assessment of solidmunicipal wasteconditions.

Theuseofthemethodsofthemultivariatedataanalysisandtheresultsinter-pretation in the formofdigital statematrixesallow todistinguishheteroge-neousobjectsofvariousgeo-mechanicalstructureandchemicalpropertiesinthe municipal solid waste structure depending on the conditions of objectformationandthegeo-environmentstayperiod.

Differentiation of solid waste landfills on the heterogeneous parts and re-searchofthegeo-mechanicalcharacteristicsofthelandfillsoilallowtodistin-guishconstantareassuitablefortheconstructionofthesolidmunicipalwasterecyclingcomplex.

66

P24.StudyofthevariationsinrecoveredpaperpropertieswiththehelpofmultivariateanalysisRuslanO.ShabievSaintPetersburgStateUniversityofIndustrialTechnologiesandDesign,Russia

Itiswellknownthatthepulpandpaperpropertiescorrelatewitheachother,but the traditional one-to-one-factor analysisdoesnot establish the exhaus-tiverelationshipbetweenthembecauseofavastnumberofdifferentproper-tiesbeingcountedandmeasured.

Thepulpandhandsheetqualityparametersoftherecoveredpaperfromdif-ferentsourceshavebeenobtainedbasedonlaboratorymeans.Therecoveredpaperismainlyrepresentedbytheusedcorrugatedcasematerialsnamelyoldcorrugatedcontainersgrade(OCC).Asmuchastenrecoveredpapersampleshavebeenwithdrawnandanalyzedforthevariationsofnineindependentre-sponses.TheapproachbeingusedintheresearchwasfullybasedonChemo-metrics inExcel – the course andbookwritten byProfessorAlexeyPomer-antsev[1].

The study found that there are two large groups in the scores plot – one isrepresentedbythebestsamplesandanotherbythesampleswhichhavebeenbeing influencedbysomeunknown factors.The loadingsplotallowed to in-terpretthepresenceofthegroups.ThestudyhasprovenapossibilityofusingPCAforanalysisofpulpandpaperproperties.

References

1.PomerantsevA.,ChemometricsinExcel,Tomsk,2014,435p.

67

Participants

68

AmigoJoseManuelUniversityofCopenhagenAssociateprofessorCopenhagen,Denmarkjmar@life.ku.dk

BabushkinValeriyLLCFOSSElectricApplicationengineerMoscow,Russiavb@foss.dk

BelikovVadimSamaraStateTechnicalUniversityPhstudentSamara,Russiamymailbox-desu@yandex.ru

BelikovaValeriaSamaraStateTechnicalUniversitySamara,Russiavalerya.belickova@yandex.ru

BogomolovAndreyartphotonicsGbmHBerlin,Germanyab@globalmodelling.com

ChertesKonstantinSamaraStateTechnicalUniversityProfessorSamara,Russiachertes2007@yandex.ru

CruzJordiEscolaUniversitàriaSalesianadeSarriàLecturerSabadell,Spainjcruz@euss.cat

DebusBrunoSaintPetersburgStateUniversityPostdocSaintPetersburg,Russiab.debus87@gmail.com

DokuchaevAlexandrSamaraStateTechnicalUniversitySoftwareengineerSamara,Russiadocuhaevrud@gmail.com

DubrovskiyDmitriyBashkirStateUniversitystudentUfa,Russiadu315@mail.ru

DuponchelLudovicLille1UniversityProfessorLille,Franceludovic.duponchel@univ-lille1.fr

EgorovaEkaterinaSaintPetersburgStateInstituteofTechnol-ogyPhDstudentSaintPetersburg,Russiakaterinka7@list.ru

ErmakovVasiliySamaraStateTechnicalUniversityHeadoflaboratorySamara,Russiawassiliy@rambler.ru

EsbensenKimKHEConsultingProfessorCopenhagen,Denmarkke@geus.dk

FilippovMaximCJSC"KoudijsMKorma"QA/QCmanagerMoscow,Russiamfilippov@kmkorma.ru

GalkinEvgeniiCSort,Ltd.HeadofR&DdepartmentBarnaul,Russiatech9@csort.ru

69

GalyaninVladislavSamaraStateTechnicalUniversityEngineerSamara,Russiav.galyanin@gmail.com

GuryanovaAlenaSamaraStateTechnicalUniversitySamara,Russiabelki.da@gmail.com

HopkinsDavidBattleCreek,USANIR.Consultant@sbcglobal.net

JanosElekSciencePortLtd.ManagerDebrecen,Hungaryelek@scienceport.hu

KabirovaLianaBashkirStateUniversitystudentUfa,RussiaKabirova.lian@yandex.ru

KalambetYuriAmpersandLtd.GeneraldirectorMoscow,Russiakalambet@ampersand.ru

KapustinDenisCSort,Ltd.EngineerBarnaul,Russiadenkapustin@csort.ru

KarpushkinEvgenyMoscowStateUniversitySeniorresearcherMoscow,Russiaeukarr@gmail.com

KhannanovMaximEnSpectrLCCHeadofdepartmentChernogolovka,Russiamk@enspectr.com

KhaydukovaMariaSaintPetersburgStateUniversityPhDstudentSaintPetersburg,Russiakhaydukova.m@gmail.com

KirsanovDmitrySaintPetersburgStateUniversityAssociateprofessorSaintPetersburg,Russiad.kirsanov@gmail.com

KosarevskayaSvetlanaSaintPetersburgStateInstituteofTechnol-ogyPhDstudentSaintPetersburg,Russiakossas13@mail.ru

KotenkoAndreySamaraStateTechnicalUniversityAssociateprofessorSamara,Russiaako1959@mail.ru

KozhukharAlexeyLLCFOSSElectricSeniorapplicationengineerMoscow,Russiaak@foss.dk

KrylovIvanMoscowStateUniversityStudentMoscow,Russiakrylov.ivan@gmail.com

KucheryavskiySergeyAalborgUniversityAssociateprofessorEsbjerg,Denmarksvkucheryavski@gmail.com

70

KurkinaVictoriaSaintPetersburgStateInstituteofTechnol-ogyAssociateprofessorSaintPetersburg,Russiavictoria.kurkina@gmail.com

LabutinTimurMoscowStateUniversitySeniorresearcherMoscow,Russiatimurla@laser.chem.msu.ru

LeardiRiccardoUniversityofGenoaAssociateprofessorGenova,Italyriclea@difar.unige.it

LykinaAnastasiaSamaraStateAerospaceUniversityStudentSamara,Russiazuum63@mail.ru

MariniFedericoSapienzaUniversityofRomeResearcherRome,Italyfederico.marini@uniroma1.it

MelentevaAnastasiiaSamaraStateTechnicalUniversityEngineerSamara,Russiamelenteva-anastasija@rambler.ru

MonakhovaYuliaSaratovStateUniversitySeniorresearcherSaratov,Russiayul-monakhova@mail.ru

MoshchenskayaElenaSamaraStateTechnicalUniversityAssistantprofessorSamara,Russialmos@rambler.ru

NechaevaNataliaMoscowStateUniversityPhDstudentMoscow,RussiaNechaeva.n.l@yandex.ru

PanchukVitalySaintPetersburgStateUniversityAssociateprofessorSaintPetersburg,Russiavitpan@mail.ru

PodchasovAntonCSort,Ltd.SoftwareengineerBarnaul,Russiapodchasov@csort.ru

PomerantsevAlexeyInstituteofChemicalPhysicsRASPrincipalresearchfellowMoscow,Russiaforecast@chph.ras.ru

PystinVitalySamaraStateTechnicalUniversityEngineerSamara,Russiavitaliy.pystin@yandex.ru

RamenskayaEkaterinaSamaraStateTechnicalUniversitySamara,Russiailjushinakatja@rambler.ru

RodionovaOxanaInstituteofChemicalPhysicsRASLeadingresearcherMoscow,Russiarcs@chph.ras.ru

RudnitskayaAlisaUniversityofAveiroResearcherAveiro,Portugalalisa@ua.pt

71

RusinovLeonSaintPetersburgStateInstituteofTechnol-ogyChiefofdepartmentSaintPetersburg,Russialrusinov@yandex.ru

SavelievAlekseySamaraStateTechnicalUniversityPhDstudentSamara,Russiaalekssaveliev@mail.ru

SavinkovMaximCSort,Ltd.CEOBarnaul,Russiasmv@csort.ru

ShabievRuslanSaintPetersburgStateUniversityofIndus-trialTechnologiesandDesignAssociateprofessorSaintPetersburg,Russiancjob@yandex.ru

SkvortsovAlexejSaintPetersburgStatePolytechnicalUni-versityHeadofdepartmentSaintPetersburg,Russiacolbug@mail.ru

SpiridonovAlexanderLLCFOSSElectricApplicationmanagerMoscow,Russiaasp@foss.dk

YarchukAnnaMoscowStateUniversitystudentMoscow,Russiaanechka1955@yandex.ru

YarkaevaYuliaBashkirStateUniversityPhDstudentUfa,Russiajulijajarkaeva05@gmail.com

ZarubinAlekseyTomskPolytechnicalUniversityAssistantprofessorTomsk,Russiazagtpuru@gmail.com

ZaytsevSergeyMoscowStateUniversityPhDstudentMoscow,Russiasergz@laser.chem.msu.ru

ZhilinSergeyAltaiStateUniversityAssociateprofessorBarnaul,Russiaszhilin@gmail.com

ZilbergRufinaBashkirStateUniversityLecturerUfa,RussiaZilbergRA@yandex.ru

ZontovYuryNationalResearchUniversityHigherSchoolofEconomicsAssistantprofessorMoscow,Russiayury.zontov@gmail.com