Jou nal o Food Composi ion and Analysis 128 (2024) 106015
A ailable online 26 Janua y 2024
0889-1575/© 2024 The Au ho (s). Published by Else ie Inc. This is an open access a icle unde he CC BY-NC-ND license (h p://c ea i ecommons.o g/licenses/by-
nc-nd/4.0/).
Yoghu s anda diza ion using eal- ime NIR p edic ion o milk a and
p o ein con en
D. Cas o-Reigía
a
,
b
, J. Ezena o
c
, M. Azkune
d
, I. Ayes a
e
, M. Os a
, J.M. Amigo
g
,
h
, I. Ga cía
b
,
M.C. O iz
a
,
*
a
Uni e sidad de Bu gos, Depa amen o de Química, Facul ad de Ciencias, Plaza Misael Ba˜
nuelos s/n, Bu gos 09001, Spain
b
Ad anced Op ical Technologies S.L. (AOTECH), Escuela Ing. de Bilbao, Plaza Ingenie o To es Que edo, 1 2º, Bilbao 48013, Spain
c
Uni e si a Ro i a i Vi gili. Chemome ics and Senso ics o Analy ical Solu ions (ChemoSens) g oup, Depa men o Analy ical Chemis y and O ganic Chemis y,
Campus Sescelades, N4 building, C/Ma cel⋅lí Domingo 1, Ta agona 43007, Spain
d
Depa men o Elec onic Technology, Enginee ing School o Bilbao, Uni e si y o he Basque Coun y (UPV/EHU), Plaza Ingenie o To es Que edo 1, 48013 Bilbao,
Spain
e
Depa men o Applied Ma hema ics, Enginee ing School o Bilbao, Uni e si y o he Basque Coun y (UPV/EHU), Plaza Ingenie o To es Que edo 1, 48013 Bilbao,
Spain
Depa amen o Applied Chemis y, Facul y o Chemis y, Uni e si y o he Basque Coun y (UPV/EHU), Manuel La dizabal 3, 20018 Donos ia, Spain
g
IKERBASQUE, Basque Socie y o he P omo ion o Science, Plaza Euskadi, 5, Bilbao 48009, Spain
h
Depa men o Analy ical Chemis y, Uni e si y o he Basque Coun y UPV/EHU, Ba io Sa iena S/N, Leioa 48940, Spain
ARTICLE INFO
Keywo ds:
Pa ial leas squa es eg ession (PLSR)
Nea -in a ed (NIR)
In-line
P oo o concep
Yoghu
Fa
P o ein
ABSTRACT
A sys em based on nea -in a ed (NIR) spec oscopy has been de eloped o he in-line con ol o he composi ion
o he milk used as aw ma e ial o yoghu p oduc ion o con ol he con en o p o ein and a in he inal
p oduc , and, he e o e, o educe a iabili y in he p oduc ion p ocess. Fi s ly, a e selec ing he app op ia e
me hod o p ep ocessing NIR da a, Pa ial Leas Squa es Reg ession models we e buil o p edic a and p o ein
con en in milk, ob aining good pe o mances. The a iance explained o y-block in p edic ion (R
2
P) was 0.99
and 0.80, while he Roo Mean Squa e E o o P edic ion (RMSEP), was 0.26 and 0.16 o a and p o ein,
espec i ely. Wi h hose models, i was possible o de e mine he a and p o ein con en s in milk in eal ime,
and he e o e, he quan i y o milk powde and c eam added in he manu ac u ing p ocess o yoghu could be
eadjus ed. The p esen ed s a egy allows he imp o emen o he homogenei y o he inal p oduc , educing he
a iabili y o he nu i ional alues in mo e han 70% wi h espec o he adi ional ecipe, and also mee he
a ge alues acco ding o yoghu p oduce s o a and p o ein con en , ha is, 10% o a and 5% o p o ein.
1. In oduc ion
Milk and i s de i a i es, such as yoghu , a e an essen ial pa o he
human die , wi h a conside able inc ease in i s global p oduc ion o e
he yea s, acco ding o he Food and Ag icul u e O ganiza ion o he
Uni ed Na ions (FAO) (FAO, 2022a). The wo ldwide es ima ed p o-
duc ion has inc eased om 466000 million kilos in 1980 o a ound
843000 million kilos in 2018, which has a conside able impac om an
economic poin o iew. Due o i s nu ien con en , nu i ional quali y,
and ene gy supply, milk is a key ood in he die a any age. I s nu i-
ional ele ance lies undamen ally in i s lipidic and p o ein ac ion ( a
cons i u es app oxima ely 3% and 4% o cow milk while p o ein is
a ound 3.5%) (FAO, 2022b). The e o e, among o he pa ame e s, a
and p o ein con en in comme cialized milk a e essen ial since i will be
decisi e in i s nu i ional alue.
Milk has been analyzed o yea s using con en ional me hods such as
he Ge be (AOAC O icial Me hod, 2000; ISO 19662:2018, 2018) and
Kjeldahl (ISO, 8968–1:2014|IDF 20–1:2014, 2014) me hods o a and
p o ein con en de e mina ion, espec i ely. The In e na ional Dai y
Fede a ion (IDF) and he In e na ional O ganiza ion o S anda diza ion
(ISO) ha e collabo a ed on all s anda diza ions, ela ing hese wo
me hods o analysis and sampling o milk and i s de i a i es o imp o e
consume p o ec ion wa an ies. These e e ence me hods ha e been
p o en ex emely use ul and con inue o play a signi ican ole in he
dai y indus y nowadays because o hei good epea abili y, among
many o he cha ac e is ics (Ma golies and Ba bano, 2018). Despi e all o
* Co esponding au ho .
E-mail add ess: [email p o ec ed] (M.C. O iz).
Con en s lis s a ailable a ScienceDi ec
Jou nal o Food Composi ion and Analysis
jou nal homepage: www.else ie .com/loca e/j ca
h ps://doi.o g/10.1016/j.j ca.2024.106015
Recei ed 15 Sep embe 2023; Recei ed in e ised o m 12 Decembe 2023; Accep ed 22 Janua y 2024
Jou nal o Food Composi ion and Analysis 128 (2024) 106015
2
ha , hei use is dec easing, as hey a e ime-consuming, and hey can
de e mine jus one pa ame e a a ime compa ed wi h some o he
ins umen al echniques. Fo ins ance, ch oma og aphic echniques can
be ano he op ion (Buz´
as e al., 2022; Delmon e e al., 2012; Sukhija and
Palmquis , 1988). Howe e , hey usually in ol e long analysis imes,
due o he necessi y o sample p epa a ion s eps like ex ac ion and
p econcen a ion, ollowed by a sepa a ion p ocess, which signi ican ly
inc eases he ime and he economic cos s o he analysis (Nielsen,
2010).
In ha sense, spec oscopic echniques can be he bes op ion
conside ing hei easy and as handling and he educ ion o analysis
ime and use o chemical p oduc s, which imply en i onmen al and
economic ad an ages. Some o he mos used spec oscopic echniques
o measu ing milk o de i ed dai y p oduc s a e isible/nea -in a ed
(Vis/NIR) spec oscopy (Ae nou s e al., 2011; Melen e a e al., 2016;
Su ko a e al., 2019), Raman spec oscopy (Mazu ek e al., 2015),
middle-in a ed (MIR) spec oscopy (Soyeu e al., 2006), and
nea -in a ed (NIR) spec oscopy (Bi an e e al., 2022; Mel sen e al.,
2012; Růˇ
ziˇ
cko ´
a and ˇ
Sus o ´
a, 2006). In his wo k, a NIR spec ome e
has been implemen ed, allowing he simul aneous and eal- ime de e -
mina ion o a and p o ein con en in milk low.
NIR spec oscopy was accep ed in 2007 as an o icial analysis
me hod by AOAC o measu e a and p o ein con en in ce ain ypes o
oods, like mea (AOAC, 2007). Rega ding milk and dai y, NIR has also
been accep ed by ISO o measu e a and p o ein con en s in milk and
dai y p oduc s (ISO 21543:2020, 2020). NIR enables non-des uc i e
measu emen s wi h no sample p epa a ion (which implies economic,
empo a y, and en i onmen al ad an ages) and can be used in-line in
e lec ance mode, allowing he de e mina ion o mul iple pa ame e s
simul aneously wi h no speci ic ins alla ion o de i ed s eam as i is
no mally done by measu ing in ansmi ance mode. In his way,
eal- ime moni o ing o milk quali y pa ame e s is achie ed wi hou
al e ing he p oduc ion line. As he NIR spec al bands a e less de ined,
and hei in e p e a ion poses an ex a di icul y, ad anced mul i a ia e
analysis me hods mus be applied o quan i y he abo emen ioned
p ope ies. In his ega d, he wo kho se eg ession me hod has been
Pa ial Leas Squa es Reg ession (PLSR) (Geladi and Kowalski, 1986;
Haaland and Thomas, 1988). Showing ha he combina ion o NIR and
chemome ics o e ed such good esul s in quan i ying p ope ies o he
milk and dai y p oduc s, he In e na ional S anda d ISO 21543:2020
“Milk P oduc s-Guidelines o he applica ion o NIR spec ome y” (ISO
21543:2020, 2020) es ablishes he pe o mance c i e ia o his ype o
analysis wi h mul i a ia e calib a ion echniques.
Se e al s udies p oposed me hods ha p edic a and p o ein con-
en in milk using NIR and PLSR (Melend e as e al., 2022; Yang e al.,
2020). These in es iga ions we e ocused on he eliable de e mina ion
o some milk a ibu es o he op imiza ion o po able de ices. In his
wo k, he de e mina ion o a and p o ein in milk samples is p oposed
o he pos e io imp o emen o e he eal- ime con ol o he yoghu
manu ac u ing p ocess.
One o he objec i es o dai y p oduce s is o manu ac u e yoghu s
wi h a inal pe cen age o 10% o a and 5% o p o ein. To do ha wi h
he usual ecipe, he quan i y o comme cial c eam (wi h a known a
con en ), milk powde (wi h a known p o ein con en ) and aw milk ha
needs o be mixed o ge a inal p oduc wi h he a o emen ioned
nu i ional alues has o be calcula ed based on he a and p o ein
con en s o he aw milk. They used he a and p o ein con en s om he
p e ious ba ches o aw milk o calcula e hose con en s, ge ing a inal
p oduc wi h conside able a iabili y. As cow-milk nu i ional alues
depend on a big a ie y o ac o s (such as ood inges ion, lac a ion
s a us, season o he yea , age, illnesses…), and as he employed milk is a
mix u e coming om many indi idual animals, nu i ional alues o he
used milk a e likely no o be simila o hose o he p e ious ba ch (FAO,
2022b). This jus i ies he necessi y o de e mining he a and p o ein
con en s in eal ime in aw milk o adjus he quan i ies o he in-
g edien s and homogenize he inal p oduc . This is, o a oid la ge
di e ences o a and p o ein con en s be ween yoghu ba ches. In his
wo k i was shown ha he implemen a ion o in-line NIR e lec ance
measu emen s in he milk supply chain leads o a subs an ial accu acy
imp o emen in he homogeniza ion o he inal p oduc . The e o e, a
s anda diza ion in he manu ac u ing p ocess o yoghu s is achie ed,
a aining a highe quali y and educing cos s.
2. Ma e ials and me hods
2.1. Milk and yoghu samples
The expe imen al wo k o his in es iga ion has been ca ied ou in a
dai y ac o y in Spain (Dulce G ado S.L). Milk supplies a i e daily o he
ac o y, coming om di e en a ms and di e en supplie s, and a e
hen s o ed in anks. As hey ha e di e en o igin, as men ioned in he
in oduc ion sec ion, he nu i ional alue o hese milk samples will
a y acco ding o ha . The expe imen al milk measu emen s we e made
in-line in his ac o y in a i e-mon h pe iod, so he g ea es possible
a iabili y be ween milk samples and hei nu i ional alues could be
explo ed and eco ded as spec a. Table S1 in he supplemen a y ma-
e ial shows he numbe o samples measu ed each day du ing he i e-
mon h pe iod. This was made o ake in o accoun he di e en condi-
ions ha could a ec he inal p oduc . The a con en anges in milk
samples be ween 2.65% and 4.06%; while he p o ein con en anges
be ween 3.17% and 3.46% (w/w).
On he o he hand, a o al o 6 yoghu samples we e manu ac u ed
based on he amoun o a and p o ein calcula ed in-line wi h he
models p oposed below. Each ba ch o yoghu implies 0.2 m
3
o milk
( ank olume). Th ee ba ches o yoghu we e manu ac u ed aking wo
samples pe ba ch and hei co esponding spec a, which implies a o al
o 6 yoghu samples, bu 0.6 m
3
o milk. By knowing he a and p o ein
con en o milk in eal ime, he quan i ies o c eam ( o adjus he
amoun o a ) and milk powde ( o adjus he he amoun o p o ein) in
he 6 yoghu s we e modi ied, since hey depend on he a and p o ein
o he aw milk. These yoghu s we e sen o an ex e nal labo a o y o
hei analysis, whe e he a con en was de e mined using he g a i-
me ic me hod and he amoun o p o ein was calcula ed using he
Kjeldahl me hod (ISO, 8968–1:2014|IDF 20–1:2014, 2014).
2.2. NIR spec oscopy
The p ocedu e was made wi h he AONIR in eg a ed solu ion o
eal- ime NIR measu men s (AOTECH, 2023), including a NIR senso
coupled o a con ol and model so wa e pla o m, which allows
eco ding he NIR di use e lec ance spec a ( om 908 o 1676 nm) o
50 milk samples in eal ime. The senso was con igu ed in such a way
ha he bes spec a we e ob ained wi h a spec um eading in e al o
one second, and 50 eadings pe spec um wi h an in eg a ion ime o
Fig. 1. Measu ing and sampling sys em o milk samples. The NIR senso (blue
cylinde ) and he ou le o sampling (indica ed wi h a ed ci cle) a e ins alled
in he pipe whe e he milk lows om he s o age ank o he pas eu iza-
ion sys em.
D. Cas o-Reigía e al.
Jou nal o Food Composi ion and Analysis 128 (2024) 106015
3
0.012 s. Each sample was measu ed in such a way ha bo h sampling
and NIR measu emen s we e pe o med simul aneously wi hou any
specialized membe . Fig. 1 ep esen s he measu emen and he sam-
pling sys em whe eas Figu e S1 in he supplemen a y ma e ial shows he
AONIR so wa e ou pu . Daily milk p oduc ion is s o ed in la ge anks,
om which he milk lows h ough pipes o he pas eu iza ion sys em.
Th ough ha low, samples a e collec ed and measu ed. Bo h he NIR
senso in Fig. 1 (blue cylinde ) and he ou le o sampling (indica ed
wi h a ed ci cle) a e ins alled in he pipe. Once he pipe is illed, a NIR
spec um is manually eco ded and immedia ely a e ha , 40 mL o
milk is aken o an ex e nal labo a o y o be analyzed wi h MILKOSCAN,
a ce i ied e e ence me hod by AOAC (AOAC, 2016) and ISO and IDF
(ISO, 9622:2013|IDF 141:2013, 2013). On he o he hand, Figu e S1
shows he ou pu o he AONIR so wa e, gi ing as an example one
spec a o milk eco ded du ing his wo k.
Wi h he 50 samples, calib a ion models we e made o model he
milk a and p o ein con en . A e wa ds, six new samples we e
measu ed and used o make yoghu s whose a and p o ein we e also
measu ed by ce i ied labo a o ies.
2.3. Mul i a ia e eg ession models
To make he eading o his pape easie , he elemen s ha a e going
o be used in i a e explained below. The ela ion be ween he collec ed
NIR spec a (X) and he co esponden esponses ( a and p o ein con-
en , y) is es ablished using a PLSR. This algo i hm builds linea com-
bina ions o X and y, maximizing hei co a iance, and inding a new se
o la en a iables (LVs) in X-block and y-block maximally ela ed o
hem (Geladi and Kowalski 1986).
PLSR is especially use ul when he p edic o s a e highly collinea o
when he numbe o p edic o s is highe han he numbe o obse a-
ions, being widely used in spec oscopic applica ions (Haaland and
Thomas 1988). Ne e heless, o build an op imalmodel, da a mus be
p ope ly p ep ocessed o educe as much unwan ed a ia ions as
possible, such as ins umen al o he mal noise, sample backg ound o
ligh sca e ing e ec s as hese e ec s a e mo e p ominen in di use
e lec ance measu emen s han in ansmi ance measu emen s. The
mos widely used p ep ocessing me hods can be di ided in
sca e -co ec ion me hods and spec al de i a i es, (Chu e al., 2022;
Mas e al., 2020; Rinnan e al., 2009; Schoo e al., 2020). Wi hin he
i s , he Mul iplica i e Sca e Co ec ion (MSC) and he S anda d
No mal Va ia e (SNV) we e he ones used in his wo k, whe eas
Sa i zky-Golay (S-G) de i a i e calcula ion was applied as he spec al
de i a i e me hod. Da a we e also mean-cen ed (MC) be o e modelling.
The SNV no maliza ion is mainly used o co ec ligh sca e ing
e ec s and changes in he op ical pa h on he NIR e lec ion spec a (Chu
e al., 2022; Rinnan e al., 2009; Schoo e al., 2020) al hough i could be
sensi i e o noisy en ies in he spec um, since i does no in ol e a leas
squa e i ing in hei pa ame e es ima ions. SNV ma hema ical de ails
can be consul ed in he e e ences indica ed abo e.
The pu pose o Mul iplica i e Sca e Co ec ion (MSC) is p ac ically
he same as SNV, his is, o emo e he e ec s o pa icle dis ibu ion and
size. Fo a da ase o med by indi idual spec a X (1 x m), i s a e age
spec um (X) is calcula ed. Nex , a linea eg ession be ween each X and
X is pe o med, ob aining b and b
0
coe icien s by leas squa es. Then, X
is co ec ed by sub ac ing b
0
and di iding by b. MSC is pe o med
assuming independence om he wa eleng h and he a ia ions in he
composi ion o samples. I is p o ed o be co ela ed wi h SNV, in ac ,
he p ep ocessing esul s o bo h me hods should be analogous (Chu
e al., 2022).
The S-G il e adjus s a polynomial o a mo ing window h ough he
wa eleng h poin s o a spec um using leas squa es, hen, he cen al
poin o he window is p edic ed using he i ed equa ion, which can be
ma hema ically de i ed be o e he p edic ion. As he equa ion does no
i pe ec ly o he da a, his il e has a smoo hing e ec apa om he
de i a ion. S-G can emo e mos o baseline in e e ences and back-
g ound noise, bu p ope selec ion o he window wid h, he de i a i e
o de and he deg ee o he polynomial mus be made. I he window
wid h is oo small, he noise is augmen ed, bu , i he di e ence wid h is
oo la ge, he spec um becomes excessi ely smoo hed, losing in o -
ma ion on he peaks o in e es (Chu e al., 2022; Rinnan e al., 2009).
To selec ing he app op ia e numbe o LVs Vene ian blinds c oss-
alida ion me hod was employed in e e y case conside ing he
o de ing o he samples, he numbe o objec s and he p esence o
eplica e samples in he da ase . The choice o he op imal numbe o
LVs o each p ep ocessing was ob ained by compa ing he Roo Mean
Squa e E o in Calib a ion (RMSEC) and he Roo Mean Squa e E o in
C oss- alida ion (RMSECV). The op imal numbe o LVs was decided by
a h eshold in he a e o change o he RMSECV be ween wo consec-
u i e anks.
Pe mu a ion es s a e ano he way o help iden i y an o e i model
as well as p o ide a p obabili y ha he gi en model is signi ican ly
di e en om one buil unde he same condi ions bu on andom da a
(Tools: Pe mu a ion Tes -Eigen ec o Resea ch Documen a ion, 2023).
These es s in ol e epea edly and andomly eo de ing he y-block, in
such a way ha he model is ebuil a e each eo de ing wi h he
a) b)
Fig. 2. a) NIR spec a o 50 samples o milk. b) P ep ocessed NIR spec a o 50 samples o milk using SNV and S-G wi h a window wid h o 11 poin s using a second-
deg ee polynomial and a 2nd de i a i e.
D. Cas o-Reigía e al.
Jou nal o Food Composi ion and Analysis 128 (2024) 106015
4
cu en modelling se ings. In his case, h ee es s we e used, he Wil-
coxon es , he signed es and he andomisa ion es . I he modeling
condi ions a e o e - i ed, hey will o en p o ide a i o andom da a
which is be e han would be expec ed. I he p- alue is g ea e ha
0.05, he model will be p obably o e i ed a a 95% con idence le el.
A e he PLSR models we e buil , o demons a e hei e aci y, he
espec i e accu acy lines ( he p edic ed con en wi h he PLSR models
e sus he ue concen a ions) we e buil . On he one hand, o he case
o p o ein, he slope is 0.9788 whe eas he in e cep is 0.0708, and
s
yx
=0.0138. On he o he hand, o he case o a , he slope is 0.9598
while he in e cep is 0.1474, and s
yx
=0.0617. Knowing ha he ideal
si ua ion o an unbiased model is when he slope is equal o one and he
in e cep ze o, he cen e s o he ellipses, ep esen he alue o he slope
and he in e cep o he models. The egion de ined wi hin he ellipse
co esponds o he con idence egion calcula ed, in his wo k, a he 95%
con idence le el.
2.4. So wa e
PLS Toolbox 8.8.1 (Wise e al., 2022) o use wi h MATLAB
(R2020b) (MATLAB, 2022) was employed o i ing he PLSR models. A
homemade p og am on MATLAB was used o calcula e he con idence
ellipses.
3. Resul s and discussion
3.1. NIR spec a o milk
The ob ained spec a a e ep esen ed in Fig. 2a. Rela ed o wa e
bands, in he milk spec a o he samples, h ee main bands we e
obse ed: one a a ound 980 nm, which co esponds o he second
o e one o he symme ic and he asymme ic s e ch o he wa e
molecule, a second one a app oxima ely 1200 nm, which co esponds
o he i s o e one o he symme ic s e ch, he bending mode, and he
asymme ic s e ch, and he hi d one, a nea ly 1450 nm, ha co e-
sponds o he i s o e one o he symme ic and asymme ic s e ch o
he wa e molecule (Weye , 2007). I can also be obse ed a small peak
a ound 1300–1350 nm ha co espond o he i s o e one o he
symme ic s e ch o he wa e molecule. In he case o he p o ein
con en , he cha ac e is ic bands in NIR will be hose associa ed wi h he
unc ional g oups ha de ine he aminoacids ha con o m he p o eins,
such as he N-H and he -COOH g oups. The e o e, he mos ele an
bands a e hose co esponding o he i s and second o e ones o N-H
s e ching (a ound 1500 nm and be ween 973 and 1020 nm, espec-
i ely). Also, he cha ac e is ic bands ha co espond o he i s N-H
o e one o he symme ic and asymme ic combina ion s e ch o
p ima y amides (1470 nm) (Weye , 2007). The band a ound 930 nm
could indica e he hi d o e one o C-H s e ch ib a ions o i-
glyce ides o he hi d o e one o he O-H bond (Ae nou s e al., 2011).
Fig. 2b shows he p ep ocessed spec a pe o ming he SNV and he
2nd de i a i e (as his is he op imal p ep ocessing o he models
employed o p edic a and p o ein con en s in milk). Wi h his ep e-
sen a ion, he mos impo an and p e iously men ioned bands a e
highligh ed, and i hey a e eally ela ed o a and p o ein con en ,
should be simila o he peaks o he loading alues ha will be ob ained
wi h he inal models used o p edic a and p o ein.
3.2. PLSR calib a ion models o a and p o ein
The igu es o me i ob ained o he PLSR models buil wi h each
conside ed p ep ocessing me hod a e summa ized in Table 1. Consid-
e ing he p e ious easoning in he Sec ion 2.2, he selec ed p e-
p ocessing me hod was SNV combined wi h S-G wi h a window wid h o
11 poin s using a second-deg ee polynomial and a 2nd de i a i e o
bo h a and p o ein. Da a we e mean-cen ed be o e modelling. The
numbe o LVs was chosen using he ene ian blinds c oss- alida ion
Table 1
PLSR models i ed o a and p o ein o each p ep ocessing. L.V., numbe o la en a iables; R
2
C, a iance explained o Y block in i ing; R
2
CV, a iance explained o Y block in c oss- alida ion; R
2
P, a iance explained
o Y block in p edic ion; RMSEC, Roo Mean Squa e E o in Calib a ion and RMSECV (%); RMSECV, Roo Mean Squa e E o in C oss-Valida ion (%); RMSEP, Roo Mean Squa e E o in P edic ion (RMSEP). p- alue is he
signi icance o he c oss- alida ed pe mu a ion es s.
p- alue
P ep ocessing me hod LV R
2
C R
2
CV R
2
P Bias CV Bias RMSEC
(%)
RMSECV
(%)
RMSEP
(%)
Pai wise Wilcoxon signed ank
es
Pai wise signed ank
es
Randomisa ion -
es
Fa SNV+MC 5 0.895 0.853 0.986 -4.44 ×10
−16
-1.99 ×10
−3
0.10 0.12 0.29 0.000 0.008 0.005
MSC+MC 4 0.883 0.851 0.986 2.22 ×10
−15
-6.81 ×10
−4
0.11 0.12 0.21 0.001 0.011 0.006
2nd de i a i e+MC 6 0.931 0.884 0.951 0.00 3.06 ×10
−3
0.08 0.10 0.40 0.000 0.000 0.005
2nd de i a i e
+SNV+MC
6 0.939 0.890 0.944 0.00 3.82 ×10
−3
0.08 0.10 0.34 0.000 0.000 0.005
*SNV+2nd de i a i e
+MC
6 0.960 0.907 0.991 -1.02×10
-14
1.42 ×10
−3
0.06 0.09 0.25 0.000 0.001 0.005
P o ein SNV+MC 6 0.826 0.720 0.785 -8.88×10
-16
-1.74 ×10
−4
0.04 0.05 0.65 0.000 0.004 0.006
MSC+MC 6 0.827 0.720 0.797 -8.88×10
-16
-2.17 ×10
−4
0.04 0.05 0.65 0.000 0.003 0.006
2nd de i a i e+MC 3 0.565 0.422 0.008 0.00 8.57 ×10
−4
0.07 0.08 0.72 0.008 0.064 0.009
2nd de i a i e +SNV
+MC
4 0.606 0.439 0.029 4.44×10
-16
-1.80 ×10
−3
0.06 0.08 0.71 0.004 0.030 0.008
*SNV+2nd de i a i e
+MC
9 0.979 0.861 0.800 0.00 1.27 ×10
−3
0.01 0.04 0.16 0.000 0.001 0.005
*
Selec ed inal models
D. Cas o-Reigía e al.
Jou nal o Food Composi ion and Analysis 128 (2024) 106015
5
p ocedu e, wi h six LVs being op imal o he a p edic ion model and
nine o he p o ein. The global pe cen age o explained a iance in he
calib a ion se is 96% o a and 98% o p o ein, while in he CV se a e
91% and 86%, espec i ely. The absence o o e i ing has been e alu-
a ed by doing h ee pe mu a ion es s (50 i e a ions) using he esiduals
in CV and all p- alues o he pe mu a ion es we e lowe ha 0.05.
Fig. 3a ep esen s he accu acy line o p o ein model and Fig. 3b
shows he co esponden loadings. Obse ing Fig. 3b, he LV wi h highe
posi i e co ela ion wi h he di use e lec ance NIR spec a a e he i s
(73% o he X-block a iance) and he second one (11% o he X-block
a iance). The i s LV is mo e ela ed wi h wa e con en since he
bands a ap oxima ely 1150 nm and 1350 nm co espond o he i s
o e one o he symme ic s e ch, he bending mode and he asym-
me ic s e ch o he wa e molecule. In he second LV, a gen le peak is
obse ed a app oxima ely 980 nm, ha , would co espond o he N-H
s e ch second o e one, su ely ela ed wi h he p esence o p o eins.
The bands co esponding o he i s and second o e ones o N-H
s e ching can be seen a ound 1500 nm and, he cha ac e is ic bands
ha co espond o he i s N-H o e one o he symme ic and asym-
me ic combina ion s e ch o p ima y amides (1470 nm) (Weye ,
2007). Besides, high nega i e loadings a e obse ed a a ound 1450 nm
in he second LV ha emphasize ha he second a iable is no ela ed
wi h wa e con en since ha wa eleng h co esponds o he i s
o e one o he symme ic and asymme ic s e ch o he wa e
molecule.
On he o he hand, Fig. 3c ep esen s he esul s o a model.
Obse ing Fig. 3d, he a loadings o he i s and second LV, explain
34% and 30% o he X-block a iance, espec i ely, being he mos
impo an ones. I appea s ha he loading weigh s ha con ibu e mos
o he PLSR models a e mainly posi i e. Tha indica es a posi i e co -
ela ion be ween a composi ion and he di use e lec ance o milk
samples.
Fig. 4 shows he con idence ellipses a a signi ican le el o 95%,
demons a ing he e aci y o bo h eg ession models by means o he
accu acy lines. As i can be seen, he con idence egion includes one o
he slope and ze o o he in e cep , espec i ely, so i can be a i med
ha he model is no biased, nei he in a p opo ional way no in a
cons an way (slope=1). Also, i can be obse ed ha he p o ein p e-
dic ion model is mo e accu a e han he he a p edic ion model,
because he e is a la ge esidual s anda d de ia ion o a (and he e-
o e, he ellipse is also la ge ).
Wi h all hese esul s, i can be ound he model buil o he de e -
mina ion o a con en had he bes pe o mance, while he a iance
explained by he p o ein model is lowe . A possible explana ion o his
lies in he size o a pa icles in milk, since i is simila o he wa e-
leng hs o he spec a (a ound 1000 nm) (Bogomolo e al., 2013), he
a) b)
c) d)
Fig. 3. a) Accu acy line o p o ein model. b) Loadings o he i s wo LVs o p o ein model. In blue he loadings o he i s LV and, in o ange he loadings o he
second LV. c) Accu acy line o a model. d) Loadings o he i s wo LVs o a model. In g een he loadings o he i s LV and, in pu ple he loadings o he
second LV.
D. Cas o-Reigía e al.
Jou nal o Food Composi ion and Analysis 128 (2024) 106015
6
a pa icles will ha e s onge capaci y o e lec ligh . On he o he
hand, since he size o p o ein micelles, like he casein p esen in he
milk, is no mally lowe han 200 nm (Rue imann and Ladisch, 1987),
he sca e ing is dependen on he wa eleng h, he e o e, he in ensi y
will dec ease wi h he ene gy o ligh .
3.3. In-line alida ion o p o ein and a con en p edic ion models
Wi h he models p oposed in he p e ious sec ion, new measu e-
men s o a and p o ein we e made on six samples o lowing milk as a
p oo o concep . The esul s ob ained o a and p o ein in milk wi h
AONIR along wi h he lab esul s using a e e ence me hod (FOSS Mil-
koscan™ FT equipmen ) can be ound in Table 2. Wi h ha da a, i o
e e ence alues o p edic ed alues (R
2
) is simila o he c oss-
alida ion p edic ions o he a model, while in he case o p o ein
he ex e nal alida ion i is a bi lowe .
3.4. Co ec ion in he yoghu manu ac u e
As he measu emen s we e made in-line and he esul s o a and
p o ein con en we e ob ained in eal ime, yoghu manu ac u ing was
made acco ding o hose esul s ins ead o using he usual ecipe, which
uses alues om p e ious ba ches. This made possible o educe he
a iabili y in he manu ac u ing p ocess and o ob ain yoghu wi h
con en s o a and p o ein ha a e mo e in acco dance wi h wha is
desi ed.
Table 3 shows he mean alues and he s anda d de ia ions o a
and p o ein con en ob ained wi h he e e ence me hod in six di e en
samples using he adi ional ecipe, bo h o milk and yoghu . As he
ela i e s anda d de ia ions (RSD) show, he a ia on o a and p o ein
con en s in yoghu is conside able.
Conside ing hese esul s, a solu ion has been p oposed based in in-
line NIR di use e lec ance and PLSR models. Once he p oblem wi h he
a ia ions in a and p o ein con en o yoghu ob ained using he
o iginal ecipe was ound, i has been conside ed ha a homogeniza ion
o he inal p oduc and i s nu i ional alues is neccesa y. Fo his
pu pose, a p oo o concep was ca ied ou using six samples. Table 4
ep esen s on he one hand, he a and p o ein con en in milk calcu-
la ed in-line using di use NIR e lec ance combined wi h he PLSR
models de eloped in Sec ion 3.2. Wi h hose esul s, he quan i ies o he
ing edien s (c eam and milk powde ) ha needed o be added o
manu ac u e yoghu s wi h he co ec con en s o a and p o ein we e
calcula ed. On he o he hand, a and p o ein con en s o he manu-
ac u ed yoghu s ( ha we e calcula ed by g a ime y and using he
kjeldahl me hod, espec i ely) a e also ep esen ed. I he mean alues
o yoghu s in Table 3 a e compa ed wi h hose in he Table 4, i can be
seen ha hey a e simila o he yogu hs, ne e heless, obse ing he
s anda d de ia ion and he RSD alues, i is clea ha in he second case,
hey a e smalle . In his way, a homogeniza ion o he inal p oduc was
achie ed, ge ing simila alues o nu i ional con en in he six yoghu
samples educing he a iabili y in a con en by 72% and by 75% in he
case o p o ein.
Wi h da a om Tables 3 and 4, an in e al-based es is applied o
decide he non-in e io i y o he mean alue o he pe cen age o a and
p o ein in ela ion o a a ge alue (as a o emen ioned, one o he ob-
jec i es o dai y p ocedu es is o manu ac u e yoghu s wi h a inal
pe cen age o 10% o a and 5% o p o ein) (O iz, 2020).
The null hypo hesis o his es is µ - T <ΔL, whe e µ is he sample
mean, T he a ge alue and, ΔL he lowe equi alence di e encial, ha
is, he di e ence allowed o asses he non-in e io i y. This non-
in e io i y es is designed o demons a e ha he mean alue ob-
ained wi h a p ocedu e ( e e ence me hod o new p ocedu e) is no
lowe han a a ge alue (10% in he case o a and 5% o p o ein). In
Fig. 4. 95% con idence le el ellipse o a and p o ein PLSR models. The ci -
cles ep esen he slope and he in e cep o he PLSR models, and he ellipse,
i s con idecnce in e al a a 95% con idence le el, in ed o p o ein and in
g een o a .
Table 2
P oo o concep . Fa and p o ein pe cen age in milk samples using a e e ence
me hod and using AONIR wi h PLSR models. RMSEP, Roo Mean Squa e E o in
P edic ion (RMSEP); R
2
P, a iance explained o y-block in p edic ion.
Sample Re e ence a
con en (%)
P edic ed a
con en wi h
AONIR (%)
Re e ence
p o ein
con en (%)
P edic ed
p o ein con en
wi h AONIR (%)
1 3.74 3.92 3.10 3.32
2 3.74 3.84 3.11 3.32
3 2.02 2.52 3.20 3.33
4 2.73 3.03 3.19 3.33
5 3.91 4.04 3.29 3.46
6 3.92 3.90 3.28 3.46
RMSEP 0.26 0.16
R
2
P 0.99 0.80
Table 3
Mean and s anda d de ia ion alues o a and p o ein con en s ob ained wi h o
he usual ecipe o yoghu s.
Sample Fa con en
in milk
P o ein
con en in
milk
Fa con en in
yoghu
P o ein con en
in yoghu
Mean (%) 4.31 3.24 8.85 4.40
S anda d
de ia ion
1.69 0.07 0.91 0.54
RSD (%) 39 2 10 12
Table 4
Fa and p o ein pe cen age in aw milk and in yogu hs o six samples o
calcula e he s anda d de ia ion in he nu i ional con en o he inal p oduc .
Sample Fa con en
in milk (%)
P o ein
con en in
milk (%)
Fa con en in
yoghu (%)
P o ein con en
in yoghu (%)
1 3.92 3.32 8.80 4.20
2 3.84 3.32 8.50 4.40
3 2.52 3.33 8.50 4.50
4 3.03 3.33 8.50 4.60
5 4.04 3.46 9.10 4.40
6 3.90 3.46 8.90 4.50
Mean (%) 3.54 3.37 8.72 4.43
S anda d
de ia ion
0.62 0.07 0.26 0.14
RSD (%) 18 2 3 3
D. Cas o-Reigía e al.
Jou nal o Food Composi ion and Analysis 128 (2024) 106015
7
bo h cases ΔL has been ixed a 15% o he a ge alue (ΔL= − 1.5 o a
and ΔL= − 0.75 o p o ein). When using his es p ocedu e, non-
in e io i y may only be asse ed i he p- alue is less han 0.05. As can
be seen in Table 5, only when he new p ocedu e is applied, he non-
in e io i y has been demons a ed.
These esul s show ha he p oo o concep , ha needed a o al
ammoun o milk o app oxima ely 0.6 m
3
, migh p o ide a p omising
applica ion o he homogeniza ion o yoghu s as a inal p oduc and
he s anda diza ion o he manu ac u ing p ocess. This can be consid-
e ed as an ini ial app oach o analyze mo e samples and e i y indus ial
a iabili y by using and implemen ing con ol cha s wi h mo e conc e e
speci ica ions.
4. Conclusions
In his wo k, a eal- ime me hodology was de eloped o educe he
manu ac u ing a iabili y o yoghu s by quan i ying a and p o ein
con en s o he aw milk used o p oduce hem. Wi h his new app oach,
which can be easily au oma ed, he nu i ional p ope ies o he inal
p oduc (yoghu ) a e homogenised, imp o ing he s anda d me hod
wi hou he necessi y o quali ied s a o analysing he samples. In
addi ion, he easibili y o using a NIR spec ome e combined wi h
PLSR o his pu pose was p o ed. The applied chemome ics s a egy
allowed he de e mina ion o a and p o ein in eal ime wi h good
accu acy, ne e heless, he calib a ion anges used in his wo k a e
educed since hey espond o eal anges o he indus ial p ocess
because he p oduc ion p ocess is al eady es ablished in an indus y and
he a iabili y o analy ical pa ame e s o sequen ial ba ches is small.
The e o e, o u u e calib a ion main enance, hese anges should be
inc eased o he maximum ex en possible, as is usually done in he
pha maceu ical indus y. In he case o milk/yoghu , i is possible o
dilu e he samples o ob ain lowe concen a ions o o concen a e hem
(adding powde ed milk o modi y he amoun o p o ein, o c eam in he
case o a ). In gene al, his could imp o e he p ecision o he calib a-
ions. The ex e nal alida ion was ca ied ou using six yoghu samples
as a p oo o concep , so u he measu emen s a e needed o a s onge
alida ion in a u u e wo k. Yoghu manu ac u ing s anda diza ion was
accomplished p o ing ha he p oposed me hod can be success ully
implemen ed in he dai y indus y since i has been possible o educe
he a iabili y be ween ba ches o yoghu by up o 70% and also mee
he a ge alues acco ding o yoghu p oduce s o a and p o ein
con en , which a e 10% o a and 5% o p o ein.
Funding
This p ojec has ecei ed unding om he Eu opean Union’s Ho i-
zon 2020 esea ch and inno a ion p og amme unde g an ag eemen
No. 824769 and om he Eu opean Union-Nex Gene a ionEU unde he
INVESTIGO p og ame. G an URV Ma í i F anqu´
es – Banco San ande
(2021PMF-BS-12).
CRediT au ho ship con ibu ion s a emen
Amigo J.M,: Concep ualiza ion, Fo mal analysis, Supe ision, Vali-
da ion, W i ing – e iew & edi ing. Os a M.: Concep ualiza ion, Fo mal
analysis, Supe ision, W i ing – e iew & edi ing. O iz M.C. C uz: Da a
cu a ion, Fo mal analysis, Supe ision, Valida ion, W i ing – e iew &
edi ing. Ga cía I.: Concep ualiza ion, Fo mal analysis, Funding acqui-
si ion, In es iga ion, Me hodology, P ojec adminis a ion, Supe ision.
Ezena o J.: Concep ualiza ion, Da a cu a ion, Fo mal analysis, W i ing
– o iginal d a , W i ing – e iew & edi ing. Cas o-Reigía D.:
Concep ualiza ion, Da a cu a ion, Fo mal analysis, In es iga ion,
W i ing – o iginal d a , W i ing – e iew & edi ing. Ayes a I.:
Concep ualiza ion, Fo mal analysis, Supe ision, W i ing – e iew &
edi ing. Azkune M.: Concep ualiza ion, Fo mal analysis, Supe ision,
W i ing – e iew & edi ing.
Decla a ion o Compe ing In e es
The au ho s decla e ha hey ha e no known compe ing inancial o
pe sonal ela ionships ha could ha e appea ed o in luence he wo k
epo ed in his pape . This p ojec e lec s he iews o he au ho , and
he Eu opean Union is no esponsible o any use ha may be made o
he in o ma ion i con ains.
Da a a ailabili y
The au ho s do no ha e pe mission o sha e da a.
Acknowledgemen s
Au ho s app ecia e he collabo a ion o Dulceg ado o p o iding he
milk samples du ing he p ojec and ALS LIFE SCIENCES GALICIA S.L.
o hei con ibu ion in he yogu h analysis using e e ence me hods.
Appendix A. Suppo ing in o ma ion
Supplemen a y da a associa ed wi h his a icle can be ound in he
online e sion a doi:10.1016/j.j ca.2024.106015.
Re e ences
Ae nou s, B., Polshin, E., Lamme yn, J., Saeys, W., 2011. Visible and nea -in a ed
spec oscopic analysis o aw milk o cow heal h moni o ing: e lec ance o
ansmi ance? J. Dai y Sci. 94 (11), 5315–5329. h ps://doi.o g/10.3168/jds.2011-
4354.
AOAC O icial Me hod 972.16. Fa , Lac ose, P o ein and Solids in Milk, (2016).
AOAC. O icial Me hod 2007.04. Fa , Mois u e and P o ein in Mea and Mea P oduc s.
Mid-In a ed Spec oscopy Me hod., (2007).
AOAC O icial Me hod 2000.18. Fa Con en o Raw and Pas eu ized Whole Milk. Ge be
Me hod by Weigh , (2004).
AOTECH. Ad anced Op ical Technologies. (2023). Re ie ed om h ps://www.ao ech.
es/. Accessed Decembe 11, 2023.
Bi an e, G., Pa el, N., Cecchina o, A., Be zaghi, P., 2022. In i ed e iew: a
comp ehensi e e iew o isible and nea -in a ed spec oscopy o p edic ing he
chemical composi ion o cheese. J. Dai y Sci. 105 (3), 1817–1836. h ps://doi.o g/
10.3168/jds.2021-20640.
Bogomolo , A., Melen e a, A., Dahm, D.J., 2013. Fa globule size e ec on isible and
sho wa e nea in a ed spec a o milk. J. In a ed Spec osc. 21 (5), 435–440.
h ps://doi.o g/10.1255/jni s.1076.
Buz´
as, H., Sz´
ekelyhidi, R., Sza ne , G., Szab´
o, K., Süle, J., Buko ics, S., Ko ´
acs, A.J.,
2022. De eloped apid and simple RP-HPLC me hod o simul aneous sepa a ion and
quan i ica ion o bo ine milk p o ein ac ions and hei gene ic a ian s. Anal.
Biochem. 658 h ps://doi.o g/10.1016/j.ab.2022.114939.
Chu, X., Huang, Y., Yun, Y.H., Bian, X., 2022. Chemome ic me hods in analy ical
spec oscopy echnology. Chemom. Me hods Anal. Spec osc. Technol. h ps://doi.
o g/10.1007/978-981-19-1625-0.
Delmon e, P., Fa din-Kia, A.R., K ame , J.K.G., Mossoba, M.M., Sidisky, L., Tybu czy, C.,
Rade , J.I., 2012. E alua ion o highly pola ionic liquid gas ch oma og aphic
column o he de e mina ion o he a y acids in milk a . J. Ch oma og . A 1233,
137–146. h ps://doi.o g/10.1016/J.CHROMA.2012.02.012.
FAO. Food and Ag icul u e o he Uni ed Na ions. Global Dai y sec o : S a us and ends.
(2022a).
Table 5
Non-in e io i y es s. Compa ison o a a ge alue o a and p o ein in yoghu s
when he con en has been ob ained wi h adi ional me hod o he new p o-
cedu e (in-line NIR-PLSR).
Compa ison p-
alue
Conclusion (alpha=5%)
Fa con en in
yoghu (%)
T adi ional me hod
e sus a ge alue
0.195 Non-in e io i y has no
been demons a ed
New p ocedu e e sus
a ge alue
0.047 Non-in e io i y has been
demons a ed
P o ein con en in
yoghu (%)
T adi ional me hod
e sus a ge alue
0.263 Non-in e io i y has no
been demons a ed
New p ocedu e e sus
a ge alue
0.013 Non-in e io i y has been
demons a ed
D. Cas o-Reigía e al.
Jou nal o Food Composi ion and Analysis 128 (2024) 106015
8
FAO. Food and Ag icul u e O ganiza ion o he Uni ed Na ions. (2022b). Ga eway o
dai y p oduc ion and p oduc s. Re ie ed om h ps://www. ao.o g/dai y-
p oduc ion-p oduc s/p oduc s/milk-composi ion/en/. Accessed June 6, 2023.
Geladi, P., Kowalski, B.R., 1986. Pa ial leas -squa es eg ession: a u o ial. Anal. Chim.
Ac a 186.
Haaland, D.M., Thomas, E.V., 1988. Pa ial leas -squa es me hods o spec al analyses.
Rela ion o o he quan i a i e calib a ion me hods and he ex ac ion o quali a i e
in o ma ion. Anal. Chem. 60 (11), 1193–1202.
ISO 19662:2018. Milk-De e mina ion o a con en -Acido-bu y ome ic (Ge be
me hod), (2018).
ISO 21543:2020. Milk and milk p oduc s-Guidelines o he applica ion o nea in a ed
spec ome y, (2020).
ISO 8968–1:2014|IDF 20–1:2014. Milk and milk p oduc s-De e mina ion o ni ogen
con en -Pa 1:Kjeldahl p inciple and c ude p o ein calcula ion, (2014).
ISO 9622:2013|IDF 141:2013. Milk and liquid milk p oduc s-Guidelines o he
applica ion o mid-in a ed spec ome y, (2013).
Ma golies, B.J., Ba bano, D.M., 2018. De e mina ion o a , p o ein, mois u e, and sal
con en o Chedda cheese using mid-in a ed ansmi ance spec oscopy. J. Dai y
Sci. 101 (2), 924–933. h ps://doi.o g/10.3168/jds.2017-13431.
Mas, C., Rubio, L., Val e de-Som, L., Sa abia, L.A., O iz, M.C., 2020. Impac o he
p e ea men o ATR-FTIR signals on he igu es o me i when PLS is used. Chemom.
In ell. Lab. Sys . 201, 104006 h ps://doi.o g/10.1016/j.chemolab.2020.104006.
MATLAB (2022). MATLAB e sion: 9.9.0 (R2020b). The Ma hwo ks Inc., Na ick,
Massachuse s.
Mazu ek, S., Szos ak, R., Czaja, T., Zachwieja, A., 2015. Analysis o milk by FT-Raman
spec oscopy. Talan a 138, 285–289. h ps://doi.o g/10.1016/j.
alan a.2015.03.024.
Melend e as, C., Fo cada, S., Fe n´
andez-s´
anchez, M.L., Fe n´
andez-colome , B., Cos a-
e n´
andez, J.M., L´
opez, A., Fe e o, F., Soldado, A., 2022. Nea -In a ed senso s o
onsi e and nonin asi e quan i ica ion o mac onu ien s in b eas milk. Senso s 22
(4), 1–12. h ps://doi.o g/10.3390/s22041311.
Melen e a, A., Galyanin, V., Sa enko a, E., Bogomolo , A., 2016. Building global models
o a and o al p o ein con en in aw milk based on his o ical spec oscopic da a in
he isible and sho -wa e nea in a ed ange. Food Chem. 203, 190–198. h ps://
doi.o g/10.1016/j. oodchem.2016.01.127.
Mel sen, A., Ha ung, E., Haeusse mann, A., 2012. Accu acy o in-line milk composi ion
analysis wi h di use e lec ance nea -in a ed spec oscopy. J. Dai y Sci. 95 (11),
6465–6476. h ps://doi.o g/10.3168/jds.2012-5388.
Nielsen, S.S. (2010). Food analysis (4 h ed.).
O iz, M.C., Sa abia, L.A., S´
anchez, M.S., He e o, A., 2020. Quali y o analy ical
measu emen s: s a is ical me hods o in e nal alida ion. Comp . Chemom. Chem.
Biochem. Da a Anal. 1, 1–52. h ps://doi.o g/10.1016/B978-0-12-409547-2.14746-
8.
Rinnan, Å., Be g, F. an den, Engelsen, S.B., 2009. Re iew o he mos common p e-
p ocessing echniques o nea -in a ed spec a. T AC - T ends Anal. Chem. 28 (10),
1201–1222. h ps://doi.o g/10.1016/j. ac.2009.07.007.
Rue imann, K.W., Ladisch, M.R., 1987. Casein micelles: s uc u e, p ope ies and
enzyma ic coagula ion. Enzym. Mic ob. Technol. 9 (10), 578–589. h ps://doi.o g/
10.1016/0141-0229(87)90109-8.
Růˇ
ziˇ
cko ´
a, J., ˇ
Sus o ´
a, K., 2006. De e mina ion o selec ed pa ame e s o quali y o he
dai y p oduc s by NIR spec oscopy. Czech J. Food Sci. 24 (6), 255–260. h ps://doi.
o g/10.17221/3322-cj s.
Schoo , M., Kappe , C., an Kollenbu g, G.H., Pos ma, G.J., an Kessel, G., Buydens, L.M.
C., Jansen, J.J., 2020. In es iga ing he need o p ep ocessing o nea -in a ed
spec oscopic da a as a unc ion o sample size. Chemom. In ell. Lab. Sys . 204,
104105 h ps://doi.o g/10.1016/j.chemolab.2020.104105.
Soyeu , H., Da denne, P., Deha eng, F., Lognay, G., Veselko, D., Ma lie , M., Be ozzi, C.,
Maye es, P., Gengle , N., 2006. Es ima ing a y acid con en in cow milk using mid-
in a ed spec ome y. J. Dai y Sci. 89 (9), 3690–3695. h ps://doi.o g/10.3168/jds.
S0022-0302(06)72409-2.
Sukhija, P.S., Palmquis , D.L., 1988. Rapid me hod o de e mina ion o o al a y acid
con en and composi ion o eeds u s and eces. J. Ag ic. Food Chem. 36 (6),
1202–1206. h ps://doi.o g/10.1021/j 00084a019.
Su ko a, A., Beliko a, V., Ki sano , D., Legin, A., Bogomolo , A., 2019. Towa ds an
op ical mul isenso sys em o dai y: global calib a ion o a analysis in
homogenized milk. Mic ochem. J. 149, 104012 h ps://doi.o g/10.1016/j.
mic oc.2019.104012.
Tools: Pe mu a ion Tes -Eigen ec o Resea ch Documen a ion Wiki. 2023. Re ie ed
om: h ps://www.wiki.eigen ec o .com/index.php? i le=Tools:_Pe mu a ion_Tes .
Accessed June 6, 2023.
Weye , L. (2007). P ac ical guide o in e p e a i e nea -in a ed spec oscopy.
Wise, B.M., Gallaghe , N.B., B o, R., Sha e , J.M., Winding, W., Koch, R. 2022. PLS
Toolbox 8.8.1. Eigen ec o Resea ch Inc.
Yang, B., Zhu, Z., Gao, M., Yan, X., Zhu, X., Guo, W., 2020. A po able de ec o on main
composi ions o aw and homogenized milk. Compu . Elec on. Ag ic. 177, 105668
h ps://doi.o g/10.1016/j.compag.2020.105668.
D. Cas o-Reigía e al.
