IsoFoodTrack: a comprehensive database and management system based on stable isotope ratio analysis for combating food fraud

F on ie s in Nu i ion 01 on ie sin.o g
IsoFoodT ack: a comp ehensi e
da abase and managemen
sys em based on s able iso ope
a io analysis o comba ing ood
aud
Ca h ineTe o
1,2, Robe Modic
3, Ma e žOg inc
3,
And ažSimčič
3, JanD ole
3, TomeE imo
2,3,
Ba ba aKo oušić Seljak
2,3 and Ni esOg inc
1,2*
1 Depa men o En i onmen al Sciences, Jože S e an Ins i u e, Ljubljana, Slo enia, 2 Jože S e an
In e na ional Pos g adua e School, Ljubljana, Slo enia, 3 Compu e Sys ems, Jože S e an Ins i u e,
Ljubljana, Slo enia
The IsoFoodT ack da abase is a comp ehensi e, scalable, and lexible pla o m
designed o manage iso opic and elemen al composi ion da a o a wide
ange o ood commodi ies. I suppo s esea ch in ood au hen ici y and
aud de ec ion by in eg a ing iso opic da a wi h ich me ada a, including
geog aphical, p oduc ion, and me hodological de ails. The da abase is buil
o scalabili y, allowing he addi ion o new commodi ies, analy ical me hods,
and me ada a ields, while ensu ing in e ope abili y wi h ex e nal da abases
h ough s anda dized o ma s and API in eg a ion. Based on he da a collec ed
in IsoFoodT ack using s a is ical, chemome ic and machine lea ning app oaches
i has a capabili y o iden i y and classi y he o igin o ood commodi ies.
IsoFoodT ack also suppo s iso ope mapping (isoscapes), p o iding spa ially
con inuous p edic ions ha enhance he de ec ion o ood aud. Rigo ous
quali y con ol measu es ensu e high da a eliabili y, and he use - iendly web
in e ace acili a es easy access and isualiza ion. Openly accessible h ough
pla o ms like Na ional Cen e o Biomedical On ology (NCBO) BioPo al,
IsoFoodT ack is posi ioned o u u e expansion and in eg a ion o open-
access da a, making i a i al ool o esea che s and egula o y agencies in
ensu ing ood au hen ici y and aceabili y.
KEYWORDS
da abase, s able iso ope a io analysis, elemen al composi ion, ood aud,
au hen ici y, in e ope abili y
1 In oduc ion
Food aud, which e e s o he economically mo i a ed adul e a ion and mislabeling o ood
p oduc s, con inues o bea majo issue o ood p oduce s as well as consume s. Among he
echniques a ailable o de ec ing aud, s able iso ope inge p in ing is leading he way in
es ablishing he au hen ici y and geog aphical o igin o ood p oduc s. This choice is based on
he ac ha he dis ibu ion o s able iso opes o ca bon (
12
C,
13
C), ni ogen (
14
N,
15
N), sul u (
32
S,
OPEN ACCESS
EDITED BY
Massimo Luca ini,
Council o Ag icul u al Resea ch and
Economics, I aly
REVIEWED BY
Yuwei Yuan,
Zhejiang Academy o Ag icul u al Sciences,
China
Oana Romina Bo o an,
Na ional Resea ch and De elopmen Ins i u e
o C yogenic and Iso opic Technologies
(ICSI), Romania
*CORRESPONDENCE
Ni es Og inc
[email p o ec ed]
RECEIVED 24 Oc obe 2024
ACCEPTED 03 Feb ua y 2025
PUBLISHED 19 Feb ua y 2025
CITATION
Te o C, Modic R, Og inc M, Simčič A,
D ole J, E imo T, Ko oušić Seljak B and
Og inc N (2025) IsoFoodT ack: a
comp ehensi e da abase and managemen
sys em based on s able iso ope a io analysis
o comba ing ood aud.
F on . Nu . 12:1516521.
doi: 10.3389/ nu .2025.1516521
COPYRIGHT
© 2025 Te o, Modic, Og inc, Simčič, D ole,
E imo , Ko oušić Seljak and Og inc. This is an
open-access a icle dis ibu ed unde he
e ms o he C ea i e Commons A ibu ion
License (CC BY). The use, dis ibu ion o
ep oduc ion in o he o ums is pe mi ed,
p o ided he o iginal au ho (s) and he
copy igh owne (s) a e c edi ed and ha he
o iginal publica ion in his jou nal is ci ed, in
acco dance wi h accep ed academic
p ac ice. No use, dis ibu ion o ep oduc ion
is pe mi ed which does no comply wi h
hese e ms.
TYPE O iginal Resea ch
PUBLISHED 19 Feb ua y 2025
DOI 10.3389/ nu .2025.1516521
Te o e al. 10.3389/ nu .2025.1516521
F on ie s in Nu i ion 02 on ie sin.o g
34
S), hyd ogen (
1
H,
2
H), and oxygen (
16
O,
18
O)
1
is in luenced by
ac iona ion p ocesses linked o local clima e, geology, and soil
cha ac e is ics (1, 2). These p ocesses esul in a ying a es o iso ope
ans e om na u al sou ces such as wa e , soil, and he a mosphe e o
plan o animal issues. Fo example, he iso ope a ios in wa e (2H/1H
and
18
O/
16
O) p o ides c i ical in o ma ion abou local p ecipi a ion,
su ace wa e , and g oundwa e , in luenced by ac o s like la i ude,
al i ude, dis ance om he sea, p ecipi a ion le els, and
e apo anspi a ion. The e i ica ion o egional o igin becomes e en
mo e obus when iso ope da a a e combined wi h elemen al
composi ion p o iles (3). Howe e , o de e mine au hen ici y, a sui able
e e ence da ase o analyzed au hen ic p oduc s is equi ed. This da ase
should include samples ep esen a i e o a wide ange o geog aphical,
seasonal, die a y, and p oduc ion condi ions. Au hen ici y is hen
assessed by compa ing he alues ound in comme cial samples wi h he
limi s es ima ed om he e e ence da ase , using a sui able s a is ical
model o e alua e he bes i . These da abases also need o
becon inuously cu a ed and kep up o da e, which is a conside able
ask, gi en he amoun o a ia ion ha needs o beincluded.
A p ime example o a well-es ablished da abase is he Eu opean
Wine Da aBank, which he Eu opean Commission has main ained o
o e 20 yea s (4). Howe e , while some da abases a e publicly a ailable,
many o he s a e no eely sha ed due o in ellec ual p ope y conce ns
and di e ences in sample p e- ea men me hods. Ne e heless, he use
o iso ope da abases is expanding; o example, he S able Iso ope Ra io
Analysis (SIRA) da abase is al eady being applied o p oduc s like
Pa ma ham, G ana Padano cheese, and Pa migiano Reggiano in I aly
(5). O he examples include he po k o igin da abase managed by he
UK’s Ag icul u e and Ho icul u e De elopmen Boa d (AHDB), he
egg da abase Kon ollie e Al e na i e Tie hal ungs o men (KAT), and
aspa agus da abases in Ge man ood con ol labs.
To add ess he gaps in he a ailabili y and accessibili y o such
da a, weha e de eloped a comp ehensi e da abase and da abase
managemen sys em (DBMS) called IsoFoodT ack.
2
This sys em
p o ides ex ensi e da a on he s able iso opes o ligh elemen s and
he elemen al composi ion o au hen ic samples om a ious ood
commodi ies such as oils, milk and dai y p oduc s, mea , spices,
u les, sea ood and ege ables. Fu he mo e, IsoFoodT ack is
designed o be in e ope able, allowing connec ion wi h o he
da abases o cen alized eposi o ies. IsoFoodT ack ep esen s a
signi ican ad ancemen o e adi ional ood da abases by
p io i izing bo h accessibili y and s anda diza ion. I inco po a es
open-access p inciples, ensu ing ha esea che s om di e se
egions can u ilize i s esou ces wi hou signi ican ba ie s.
1 Measu emen s o he s able iso ope a ios o ligh elemen s a e exp essed
in he δ-no a ion in ‰ acco ding o he equa ion: δ
i
E = (R(
i
E/
j
E)
sample
/
R(
i
E/
j
E)
s anda d
)– 1, whe e i s ands o he highes and j o he lowes a omic
mass numbe o he elemen E (C, N, O, S), and R is he iso ope a io be ween
he hea ie and he ligh e iso ope o he elemen (
2
H/
1
H,
13
C/
12
C,
15
N/
14
N,
18
O/
16
O,
34
S/
32
S) in he sample o s anda d. The δ
13
C alues a e exp essed ela i e
o V-PDB (Vienna-Pee Dee Belemni e) s anda d, δ
15
N alues ela i e o AIR,
δ34S alues ela i e o V-CDT (Vienna-Canyon Diablo T oili e) s anda d and he
δ
2
H and δ
18
O alues ela i e o he VSMOW (Vienna-S anda d Mean Ocean
Wa e ) s anda d.
2 h p://iso ood ack.ijs.si
Addi ionally, he da abase in eg a es s anda dized me ada a
p o ocols and ha monized da a en y o ma s, which s eamline
c oss-s udy compa isons and enhance ep oducibili y.
Addi ionally, he e is a g owing mo emen owa d c ea ing a
cen alized eposi o y o iso opic da a, as p oposed by Pauli e al.
(6), wi h he de elopmen o IsoBank. IsoBank aims o unc ion
simila ly o GenBank in he ield o gene ics, se ing as bo h an
agg ega o and eposi o y o open-access iso ope da a. IsoBank is
designed as a gene al-pu pose eposi o y o s able iso ope da a
ac oss all disciplines. I suppo s he s o age and e ie al o
iso ope measu emen s i espec i e o hei con ex . I se es a
b oad esea ch communi y, including ields like ecology, geology,
a cheology, and biology, among o he s. This esou ce will p omo e
in e disciplina y esea ch, acili a e da a-sha ing, and p o ide
aluable educa ional oppo uni ies by o e ing eal-wo ld iso opic
da a o s uden s and esea che s alike. On he o he hand,
IsoFoodT ack da abase se es as a specialized da abase aimed a
p ac ical applica ions in ood aceabili y and au hen ici y
e i ica ion, p io i izing unc ionali y ailo ed o i s speci ic use
case. I s scope is na owe , a ge ing applica ions in ood science,
ag icul u e, and egula o y amewo ks.
In his pape , wep esen he IsoFoodT ack da abase as he i s
e o o o ganize open-access s able iso ope da a o ood esea ch. I
is o ganized in di e en sec ions including: da abase design, me hods
and echnical aspec s. Sec ion 5 de ails he alida ion o IsoFoodT ack,
demons a ing i s p ac ical applica ion, while sec ion 6 deals wi h
da abase cu a ion and a ailabili y. Finally, sec ion 7 concludes he
pape by discussing key achie emen s and con ibu ions.
2 Da abase design
The design o he IsoFoodT ack da abase is c ucial o ensu ing
he e ec i e managemen o iso opic and elemen al composi ion da a
o a ious ood commodi ies. The da abase was de eloped wi h a
ocus on scalabili y, lexibili y, and da a in eg i y, enabling esea che s
o s o e, e ie e, and analyze s able iso ope da a in a s uc u ed and
e icien manne . This sec ion ou lines he key aspec s o he da abase
design, including he da a model, schema design, ela ionships
be ween en i ies, and conside a ions o main aining accu acy
and pe o mance.
2.1 Da a model
The IsoFoodT ack da abase ollows a ela ional da abase model,
which is well-sui ed o managing s uc u ed da a wi h clea ly de ined
ela ionships be ween en i ies. The ela ional model enables e icien
que ying o da a, as well as main aining consis ency and da a in eg i y
h ough he use o p ima y and o eign key cons ain s. This design
ensu es ha all da a poin s, including iso opic a ios, elemen al
composi ions, and me ada a abou samples, a e p ope ly linked o
hei ele an en i ies.
The s uc u e o IsoFoodT ack is p esen ed in Figu e1.
The main en i ies in he da abase include:
• Samples: ep esen ing he physical ood samples analyzed o
iso ope and elemen al composi ion. This also includes he ype o
Te o e al. 10.3389/ nu .2025.1516521
F on ie s in Nu i ion 03 on ie sin.o g
sample, sou ce o sample (au hen ic, comme cial), da e o
sampling and compound analysed: bulk sample ( eeze-d y o
liquid), sample wa e , ex ac ed componen s, a y acid.
• Me ada a: he me ada a co e wo ca ego ies: (i) essen ial
me ada a, desc ibing e e y da a eco d, and (ii) iso ope-speci ic
me ada a. The essen ial me ada a includes geog aphical da a:
s o ing in o ma ion abou he geog aphical loca ion o each
sample, including de ails such as la i ude, longi ude, al i ude,
dis ance om he sea. In addi ion, he da a on yea ly a e age
amoun o p ecipi a ion, a e age empe a u e o he loca ion,
geology and pedology a e also included.
P oduc ion da a: cap u ing de ails abou he p oduc ion and
p ocessing o he ood samples, such as yea o p oduc ion, ype
o ma e ial (au hen ic, comme cial), a ming p ac ices, seasonal
in o ma ion, and p oduc ion me hods (e.g., o ganic o
con en ional; i known).
The iso ope-speci ic me ada a includes e e ence ma e ials used o
no malize he da a. S able iso ope da a a e p oduced in a wide
ange o esea ch and comme cial labo a o ies. Al hough he
me hods by which he majo i y o da a, i.e., bulk ca bon (δ13C)
and ni ogen (δ
15
N) s able iso ope alues, a e s anda dized,
labo a o ies o en use sligh ly di e en p o ocols and di e en
labo a o y e e ence ma e ials o no malize da a o in e na ionally
accep ed scales. O he iso opes (e.g., δ
2
H and δ
18
O) ha e mo e
undamen al issues associa ed wi h compa abili y o
measu emen s (7). Hyd ogen in an exchangeable posi ion (e.g.,
when bound o oxygen in a hyd oxyl g oup, as in p o einaceous
ma e ial) will exchange wi h a mosphe ic wa e apo , leading o
po en ially e oneous esul s unless con olled. Thus, o ensu e
da a obus ness, quali y and use con idence, all pe inen
analy ical in o ma ion o each piece o da a is eco ded.
IsoFoodT ack me ada a includes sample p e ea men me hods
(e.g., lipid ex ac ion), e e ence ma e ials used, ype o
no maliza ion (one, wo, mul i-poin no maliza ion) and
ins umen a ion used.
• Iso ope da a: s o ing de ailed in o ma ion abou he iso opic
composi ion o each sample, including a ios o iso opes such as
δ2H, δ13C, δ15N, δ18O and δ34S (in ‰).
• Elemen al da a: cap u ing elemen al concen a ions o key
elemen s: B, Na, Mg, Al, P, S, K, Ca, V, C , Mn, Fe, Co, Cu, Zn, As,
Se, Rb, S ., Mo, Cd, Cs, Ba, Hg, Pb.
2.2 Scalabili y and in e ope abili y
The design o IsoFoodT ack an icipa es he con inuous
expansion o he da abase as new samples a e collec ed and
analyzed. As such, he da abase a chi ec u e is scalable, wi h he
abili y o accommoda e addi ional ables o new ood
commodi ies o analy ical me hods such as compound
speci ic analysis.
3 Me hods
The de elopmen o he IsoFoodT ack da abase in ol ed se e al
key s ages, including he collec ion and p epa a ion o au hen ic ood
samples, he analysis o iso opic and elemen al composi ions, he
o ganiza ion and managemen o da a, and he alida ion o he
da abase o p ac ical applica ions in ood au hen ici y es ing. This
sec ion desc ibes he me hodology employed o c ea e and cu a e he
IsoFoodT ack da abase.
3.1 Sample collec ion and p epa a ion
Sample collec ion ep esen s a c ucial s ep in he o ma ion o he
IsoFoodT ack da abase, and in o de o ensu e he accu acy o a ood
au hen ici y da abase, au hen ic samples mus beused. Ideally, samples
should becollec ed om p ima y p oduce s by impa ial collec o s o
ensu e aceabili y and au hen ici y. In compa ison, e ail samples a e
less eliable due o possible con amina ion in he supply chain. When
c ea ing he da abase, sample size and a ie y a e also impo an and
should e lec na u al a ia ions, e.g., geog aphy, b eed, and clima e.
Addi ionally, he da abase should be alida ed o i s in ended use, and
s a is ical analysis should beconside ed when de e mining sample
size. The selec ion o e e ence da a om he da abase is c ucial and
should bele o expe s only (8, 9).
In ou case, he sampling ollows an app op ia e p o ocol
de eloped o a ious applica ions o mi iga e po en ial biases
caused by he o e ep esen a ion o speci ic egions. This p o ocol
is based on ou p io expe ience and expe knowledge. To ensu e
a obus and ep esen a i e da ase , samples we e collec ed using
he ollowing c i e ia:
FIGURE1
S uc u e o he IsoFoodT ack da abase.
Te o e al. 10.3389/ nu .2025.1516521
F on ie s in Nu i ion 04 on ie sin.o g
Geog aphical di e si y: samples we e sou ced om a wide ange o
geog aphical loca ions, including di e en la i udes, al i udes, and
clima ic zones. This ensu es ha he da abase cap u es he na u al
a ia ion in iso opic and elemen al signa u es ha a ise om
en i onmen al ac o s such as local p ecipi a ion, soil ypes, and
empe a u e. In case o Slo enia, o accoun o na u al a iabili y,
sampling was designed o e lec Slo enia’s egionaliza ion, which is
di ided in o ou dis inc geog aphical egions: Dina ic,
Medi e anean, Alpine, and Pannonian.
Seasonal and empo al a ia ion: samples we e collec ed o e
mul iple g owing seasons and ha es pe iods o accoun o seasonal
changes in iso opic and elemen al composi ion. These a ia ions can
bein luenced by ac o s such as p ecipi a ion le els and empe a u e
luc ua ions h oughou he yea .
P oduc ion me hods: bo h con en ional and o ganic ood
p oduc ion me hods we e ep esen ed in he sample se .
I is also use ul o unde s and he p oduc ion densi y o a oods u
and he numbe o ele an au hen ic samples. Fo example, he
Slo enian wine da abase, es ablished in 1996, con ains 25 au hen ic
wine samples co e ing a ious geog aphical egions and a ie ies. The
da abase is also included in EU Wine Da abank. Ano he example is
ela ed o e i ica ion o co ec labeling o selec ed ui s and ege ables
on Slo enian ma ke including s awbe ies, che ies, aspa agous
apples, kaki and ga lic. Au hen ic samples a e p o ided annually by he
egional uni s o he Adminis a ion o he Republic o Slo enia o
Food Sa e y, Ve e ina y, and Plan P o ec ion om p oduce s om
a ious geog aphical p oduc ion a eas. This esea ch began in 2018 and
equi es a leas 30 au hen ic samples co e ing ou geog aphical egions.
Once collec ed, each sample was ca e ully labeled wi h me ada a,
including in o ma ion on he geog aphic o igin, da e o collec ion,
ood ype, and any ele an p oduc ion de ails. Samples we e hen
p epa ed o iso opic and elemen al analysis acco ding o s anda dized
ope a ional p o ocols, ensu ing consis ency in he ea men o
all samples.
Al hough IsoFoodT ack has been designed o co e only a limi ed
a ea, such as Slo enia, i is c ucial o ecognize he limi a ions inhe en
in egions wi h spa se da a a ailabili y on a global scale. Low-da a
egions may exhibi incomple e co e age, which can limi he
obus ness o iso opic analysis in hose a eas. To add ess his,
IsoFoodT ack could adop he ollowing wo s a egies o enhance i s
global applicabili y: encou aging con ibu ions om o he esea che s,
ini ia i es ha can ill da a gaps and build egional da ase s and
le e aging a i icial in elligence (machine lea ning) o p edic iso opic
baselines in low-da a egions, aking in o accoun app op ia e
unce ain y me ics.
3.2 Iso opic and elemen al analysis
The analysis o s able iso ope a ios and elemen al composi ion
was conduc ed using p ecise and well-es ablished echniques. The wo
main analy ical me hods used we e Iso ope Ra io Mass Spec ome y
(IRMS) and Induc i ely Coupled Plasma Mass Spec ome y
(ICP-MS) o elemen al p o iling.
3.2.1 Iso opic analysis
The e a e only a ew excep ions whe e no sample ea men is
needed, such as de e mining he δ18O alue o wa e in ood and he
δ
2
H, δ
13
C, and δ
18
O alues o oli e oil. Howe e , in mos cases, sample
ea men is equi ed since i pe mi s he isola ion o componen s ha
ha e a s onge geog aphical inge p in han he bulk sample and
wi h less in e e ence. Fo example, samples con aining lipids (e.g.,
mea , ish, milk, and cheese) a e usually de a ed because a has a
di e en C and H composi ion om he o he ood cons i uen s, and
he e o e, i s a iable quan i y can a ec he o e all iso opic signa u e.
In he IsoFoodT ack samples, de a ing was pe o med using a
mix u e o pe oleum e he and die hyl e he (2:1 / ). Be o e
analysis, all ac ions we e eeze-d ied and s o ed a
oom empe a u e.
Iso ope a ios o hyd ogen (
1
H/
2
H), ca bon (
12
C/
13
C), ni ogen
(14N/15N), oxygen (16O/18O), and sul u (32S/34S) we e measu ed using
an iso ope a io mass spec ome e (IRMS) wi h di e en p epa a ion
sys ems. The solid and liquid samples a e measu ed by elemen al
analyse coupled o IRMS (EA/IRMS), 1H/2H and 16O/18O in o ganic
ma ices wi h a TC/EA py olyse coupled o IRMS, while 1H/2H and
16
O/
18
O in wa e in ood samples is de e mined wi h Mul iFlow Bio
equilib a ion sys em connec ed o IRMS.
Valida ion o a da abase includes he da a wi hin i and i s
abili y o comple e he ole o which i was designed. All da a used
o c ea e he da abase mus be alida ed, i.e., eliable, and all
measu emen s mus ollow he p o ocol sugges ed by Sk zypek
e al. (10). The alida ion es s highligh ed in he manusc ip we e
ins umen al in assessing i s eliabili y. These es s e ealed
occasional alse posi i es and misclassi ica ions du ing bi a ia e
e alua ions, pa icula ly in ood samples wi h iso opic
composi ions nea bounda y h esholds. Fo example, oods
sou ced om egions wi h simila clima ic and en i onmen al
condi ions exhibi ed o e lapping iso opic signa u es. To add ess
hese issues, enhanced mul i a ia e analyses we e implemen ed o
imp o e classi ica ion accu acy, educing alse-posi i e a es o
below 5% in mos ca ego ies.
Fu he , i , upon e-analysis o he samples, da a ha a e
consis en wi h he ini ial “ou lie ” da a a e eco ded, u he
in es iga ion a e unde aken o de e mine he unde lying cause.
Typically, ou lie s a e due o pa icula and unique echnological o
geog aphical issues, such as a pa icula mic oclima e o
echnological choice. In his case, u he in es iga ions a e ca ied
ou o unde s and i he ou lie s belong o ano he popula ion o
da a o i hey a e jus “ou lie s” alling in he pe cen age o e o o
he chosen con idence le el ( o example, 5% o 95%
con idence le el).
I is s ongly ecommended ha labo a o ies a e acc edi ed o
ISO17025 o can demons a e ha hey ha e equi alen quali y
con ol sys ems. This is speci ically equi ed i weuse he da abank
o e i y he au hen ici y o comme cial samples o ood con ol
pu poses. Acco ding o he no m EN ISO/IEC 17025, he es
esul o an analy ical measu emen has o be s a ed wi h an
es ima e o i s unce ain y, o example, when unce ain y a ec s
compliance wi h an au hen ici y limi . Measu emen unce ain y
is usually epo ed in he e e ence me hods (in he case o o icial
and alida ed me hods), o i can bees ima ed using di e en
me hods. Dunn e  al. (11) ecen ly published guidance o
calcula ing measu emen unce ain y o s able iso ope a io del a
alues. A u he equi emen o acc edi a ion is ha labo a o y
mus pa icipa e in p o iciency es ing ha comply wi h he ISO/
IUPAC/AOAC In e na ional Ha monized P o ocol o P o iciency
Te o e al. 10.3389/ nu .2025.1516521
F on ie s in Nu i ion 05 on ie sin.o g
Tes ing o analy ical labo a o ies. In ou case his in ol es
pa icipa ion in he Food Analysis using Iso opic Techniques
P o iciency Tes ing Scheme (FIT PTS), o ganized by Eu o ins
Scien i ic h ee imes pe yea and includes samples om a ious
ood commodi ies.
3.2.2 Elemen al analysis
The elemen al composi ion o each sample was analyzed using
Induc i ely Coupled Plasma Mass Spec ome y (ICP-MS), which
is highly sensi i e o ace elemen s. The elemen al composi ion in
ou samples was pe o med on iple quad upole induc i ely
coupled plasma mass spec ome e , QQQ-ICP-MS (Agilen , USA).
P io o analysis, he samples we e diges ed using app op ia e
chemical me hods (e.g., acid diges ion) o ensu e he accu a e
quan i ica ion o elemen al concen a ions. De ec ion limi s
included in he able we e calcula ed based on h ee s anda d
de ia ions o blank measu emen s. A mo e de ailed desc ip ion o
he op imiza ion o he me hod o elemen al analysis o ui s and
ege ables can be ound in (12). The elemen al da a we e eco ded
as concen a ions (mg/kg o ppm) o key elemen s ha a e ele an
o dis inguishing di e en geog aphical o igins and
p oduc ion me hods.
Bo h iso opic and elemen al da a a e s o ed in he IsoFoodT ack
da abase, alongside he associa ed sample me ada a, o allow o
comp ehensi e compa a i e analyses. In IsoFoodT ack, iso opic and
elemen al composi ion da a a e ep esen ed as single-
poin measu emen s.
4 Technical aspec s
The cons uc ion o he da abase in ol es h ee main phases:
• Phase I: he da abase s uc u e de ini ion. The p o ocol o da a
p epa a ion was e ol ed o minimize he labo in ol ed in
popula ing he da abase. The da abase also includes me ada a
ha a e impo an o he u he e alua ion o he da a.
• Phase II: illing he da abase wi h da a gained du ing he
p ojec e olu ion.
• Phase III: de elopmen o ou ines/que ies o ex ac ing da a
om he da abase.
The undamen al equi emen s o IsoFoodT ack included (i)
he unde lying da abase and (ii) he applica ion laye
(web applica ion).
4.1 Unde lying da abase
The ollowing poin s we e conside ed in he
unde lying da abase:
Da abase pla o m: he IsoFoodT ack da abase was
implemen ed using Pos g eSQL, an open-sou ce ela ional da abase
sys em known o i s obus ness, suppo o complex que ies, and
abili y o handle la ge da ase s. The speci ic echnology was chosen
o i s eliabili y and syne gy wi h o he web echnologies.
Pos g eSQL also enables he s o age o semi-s uc u ed da a
when necessa y.
Da a impo and expo : bulk da a impo and expo we e
handled using py hon sc ip s wi h Pos g eSQL connec o s. Da a en y
was s eamlined by impo ing he iso opic and elemen al da a om
excel ile.
Unde lying da abase schema is lexible o mi iga e he need o
u he modi ica ion o accommoda e g owing me ada a o analy ical
esul s equi emen s.
A de ailed isualiza ion o he da abase’s s uc u e is p o ided in
he Supplemen a y Figu e1. This isualiza ion includes an En i y-
Rela ionship (ER) diag am ha maps ou he da abase’s ables, he
ela ionships be ween hese ables, and he a ibu es ha de ine each
en i y. The diag am iden i ies p ima y keys, o eign keys, and a ious
cons ain s, p o iding a ho ough o e iew o how da a poin s a e
linked, e e enced, and main ained ac oss he en i e
IsoFoodT ack ecosys em.
4.2 Web applica ion
The web applica ion is a use - iendly in e ace o in e ac ing and
accessing he da a om he IsoFoodT ack da abase. The IsoFoodT ack
use in e ace was buil using Django (a Py hon-based web amewo k),
which allows o in eg a ion wi h he Pos g eSQL da abase. The
landing page (Figu e2) is designed o allow use s o access he da a
quickly. All in o ma ion is o ganized in o ca ego ies, and when
selec ing a ca ego y (iso ope, elemen al composi ion), a abula display
is p esen ed wi h he ele an da a (Figu e3). The da a o elemen s a e
g ouped in o mac o, essen ial, en i onmen al, geological, and
oxic ca ego ies.
The ollowing aspec s we e conside ed o he applica ion:
• Secu i y and Audi ing (c ea ed, upda ed): The da abase is
accessible o he public only ia he web applica ion. Once all da a
is published, access will beex ended o in e es ed use s.
• Con igu a ion o me ada a and esul s: Key in o ma ion, such as
column names, da a ype, uni s, loca ion da a and ca ego iza ion,
a e included in he da abase (Figu e3).
• Re iew p ocess: Be o e any new esul s a e added o he da abase,
hey a e e iewed by he adminis a o and expe s.
• Da a Visualiza ion: Django’s ORM (Objec -Rela ional Mapping)
was used o que y he da abase and display he esul s in a use -
iendly o ma . Visualiza ion o da a was done using an
in e ac i e map wi h da a poin s ep esen ing iso ope
measu emen s on di e en ood i ems om a ious places
a ound he wo ld (Figu e4). Each da a poin is clickable and
con ains ele an in o ma ion o easie compa ison.
4.3 In e ope abili y
Le e aging he da a in e ope abili y wi hin he IsoFoodT ack
da abase, an API (Applica ion P og amming In e ace) can
beimplemen ed o allow o he sys ems o in e ac seamlessly wi h
IsoFoodT ack. This pe mi s ex e nal sys ems o esea che s o que y he
da abase and e ie e da a in s anda dized o ma s such as JSON o
CSV. The use o s anda dized da a o ma s and me ada a con en ions
ensu es compa ibili y, acili a ing da a exchange and c oss- e e encing
wi h o he iso ope da abases o cen alized eposi o ies, like IsoBank.

Te o e al. 10.3389/ nu .2025.1516521
F on ie s in Nu i ion 06 on ie sin.o g
FIGURE2
In e ace o IsoFoodT ack da abase.
FIGURE3
The abula display o a ca ego y.
Te o e al. 10.3389/ nu .2025.1516521
F on ie s in Nu i ion 07 on ie sin.o g
Fu he , he me ada a was selec ed om he ISO-FOOD on ology (13),
which desc ibes iso opic measu emen s wi h all o he necessa y
in o ma ion equi ed o u u e analysis. The on ology is linked wi h
s anda d on ologies, such as Uni s o Measu emen , Food, Nu ien and
Bibliog aphic On ologies.
F om a highe -le el pe spec i e, he sys em amewo k and
communica ion be ween he web applica ion and i s backend
p ocesses is p esen ed in Figu e5.
Backend:
• Pos g eSQL Da abase: he cen al s o age o he sys em,
con aining he collec ed and s uc u ed da a.
• Django (Da a P ocessing): ac s as he applica ion laye o da a
p ocessing and logic implemen a ion, b idging he da abase and
he nex laye .
• Node.js Se e : se es as he middlewa e o API se e ,
acili a ing communica ion be ween he backend and
he on end.
F on end:
• Boo s ap: amewo k o IsoFoodT ack iews.
• Lea le js: API ha se es map iew.
• Jque y: a ja asc ip lib a y.
5 Applica ion and alida ion
Valida ion o he IsoFoodT ack da abase was an essen ial s ep o
ensu e i s p ac ical applica ion in ood au hen ica ion and aud
de ec ion. The alida ion p ocess in ol ed se e al s ages:
5.1 Re e ence da ase alida ion
In o de o eliably de e mine au hen ici y, he iso opic da a o he
es samples mus be compa ed wi h he da abank. The mos
s aigh o wa d and s ill he mos ecognized app oach is ha o
uni a ia e da a e alua ion, based on he a i hme ic mean, median,
s anda d de ia ion and au hen ici y limi s conside ing he S uden ’s
-dis ibu ion. These me ics enable use s o unde s and he deg ee o
he e ogenei y wi hin a gi en egion and enhance he eliabili y o
o igin e i ica ion. A 95% con idence le el is conside ed app op ia e
o comme cial samples, which a e p oduced in la ge ba ches and
should ha e s able iso ope alues close o he mean alues o he
au hen ic ma e ials. The es esul can beclea , in e ms o ue o
alse, bu also suspicious o unlikely, o example, when he e e ence
da abank is no obus enough o beconside ed eliable. The mos
e icien app oach is o c ea e yea ly da abases, pa icula ly o
ege able and ui commodi ies ha exhibi signi ican a iabili y in
ha es and p oduc ion om yea o yea .
FIGURE4
In e ac i e wo ld map wi h he da a.
Te o e al. 10.3389/ nu .2025.1516521
F on ie s in Nu i ion 08 on ie sin.o g
Use s should p ima ily ely on agg ega ed egional e e ence alues
o compa ison o accoun o inhe en a iabili y wi hin he da ase . In
bo de line cases, addi ional analyses such as examining seconda y
iso opes o inco po a ing ex e nal me ada a (e.g., ace elemen s o
supply chain in o ma ion) a e ecommended. Fo samples classi ied as
“suspicious,” use s a e ad ised o conduc u he in es iga ions,
including e-analysis o consul a ion wi h expe s.
5.2 S a is ical analysis
Chemome ic me hods o mul i a ia e da a analysis help sepa a e
in o ma ion om noise, unco e hidden co ela ions, and isually
ep esen hem. The e a e h ee main app oaches: (1) explo a i e
analysis, (2) classi ica ion, and (3) calib a ion. The choice o me hod
depends on he p oblem and expe imen al da a (14). Fo example,
p incipal componen analysis (PCA) is commonly used ini ially o
dimensionali y educ ion, highligh ing he mos ep esen a i e ea u es
wi h minimal in o ma ion loss and gene a ing new a iables called
p incipal componen s (15). Howe e , PCA does no conside g oup
membe ship, so chemome ic me hods a e used o classi ica ion and
class modeling when ocusing on p oduc o igin. Classi ica ion,
synonymous wi h disc iminan me hods, assigns objec s o p ede ined
classes using echniques like linea disc iminan analysis (LDA),
k-nea es neighbo s, pa ial leas squa es-disc iminan analysis (PLS-
DA), and a i icial neu al ne wo ks (ANN) (16).
Linea disc iminan analysis, one o he simples classi ie s, equi es
a su icien a io (≥3) be ween samples and a iables and s uggles wi h
highly collinea da a common in chemis y (16). PLS-DA add esses
hese issues, c ea ing a linea model s a is ically equi alen o LDA’s
solu ion bu o e coming minimum sample- o- a iable a io and
collinea i y p oblems (16). O hogonal pa ial leas squa es-disc iminan
analysis (OPLS-DA), a modi ica ion o PLS-DA, enhances in e p e abili y
by sepa a ing p edic i e a iance om non-p edic i e a iance (17).
Model pe o mance is e alua ed by explained a ia ion (R2X o PCA
and R2Y o OPLS-DA) and p edic i e abili y (Q2), wi h in e nal
se en old c oss- alida ion minimizing o e i ing. OPLS-DA p edic ion
pe o mance is measu ed by sensi i i y ( ue posi i es) and speci ici y
( ue nega i es) (18). Disc iminan ma ke s a e selec ed by Va iable
Impo ance in he P ojec ion (VIP) alues, wi h a h eshold o one.
Class modeling, a he han disc iminan analysis, is o en
ecommended o con i m a sample’s egional o igin due o possible
FIGURE5
Sys em a chi ec u e o he IsoFoodT ack amewo k.
Te o e al. 10.3389/ nu .2025.1516521
F on ie s in Nu i ion 09 on ie sin.o g
biases in one-class classi ica ion p oblems. So independen modeling
o class analogy (SIMCA) is a s anda d me hod in chemome ics o such
asks (19, 20).
5.3 C oss- alida ion
A c oss- alida ion app oach was used o e alua e he obus ness o
he IsoFoodT ack da abase. This in ol ed spli ing he da ase in o
aining and es ing se s, whe e he aining se was used o build a
p edic i e model, and he es ing se was used o e alua e he accu acy
o he p edic ions. High p edic i e accu acy indica ed he e ec i eness
o he da abase in iden i ying ood aud and e i ying he geog aphical
o igin o samples.
5.4 P ac ical applicabili y
Finally, he p ac ical applica ion o he IsoFoodT ack da abase was
demons a ed by analyzing a se o comme cial ood samples and
compa ing hei iso opic and elemen al p o iles agains he e e ence
da ase . The esul s con i med he abili y o he da abase o de ec
disc epancies in geog aphical o igin and p oduc ion claims, he eby
alida ing i s u ili y as a ool o ensu ing ood au hen ici y. The Slo enian
s udies include he use o di e en chemome ic app oaches o e i ying
he geog aphical egion o di e en commodi ies. Fo example,
wein es iga ed he possibili y o de e mining he geog aphical o igin o
milk and dai y p oduc s. Using linea disc iminan analysis,
disc imina ion and speci ica ion o goa , cow and sheep milk and cheese
was possible (21). Mo eo e , he exis ing da abase o au hen ic Slo enian
cow milk was used o e i y he co ec assignmen o egional
p o enance and decla a ion o o igin. By applying disc iminan analysis,
he abili y o disc imina e be ween geog aphic egions was only possible
when da a we e o ganized by yea and season as a esul o di e en
eeding egimes. Based on he da a, a disc imina ion model was
de eloped o di e en ia e milk om Eu opean milk p oduced in
Slo enia e icien ly. Slo enian milk was s a is ically dis inguishable om
all o he milk, whe e he mos impo an pa ame e s we e δ18O, S ., K
and Ca. Comme cial samples labeled as “Slo enian milk” we e con i med
and classi ied as being au hen ic (22).
Despi e he ac ha he Slo enian u les sha ed some simila
cha ac e is ics wi h he samples o igina ing om o he coun ies,
di e ences in he elemen concen a ions sugges ha espec i e u le
species may espond selec i ely o nu ien s om a speci ic soil ype
unde en i onmen al and soil condi ions. C oss- alida ion esul ed in a
77% co ec classi ica ion a e o de e mining he geog aphical o igin
and a 74% co ec classi ica ion a e o disc imina ing be ween species.
The c i ical pa ame e s o geog aphical o igin disc imina ions we e S .,
Ba, V, Pb, Ni, C , Ba/Ca and S ./Ca a ios, while om s able iso opes δ18O
and δ13C alues a e he mos impo an (23).
Disc iminan and class-modeling me hods ha e also been applied o
assess he geog aphical classi ica ion and au hen ica ion o selec ed ui s
and ege ables, including s awbe ies, che ies, apples, kaki, aspa agus
and ga lic, using s able iso opes o ligh elemen s and elemen al
composi ion o samples ha es ed be ween 2018 and 2020. A good
geog aphical classi ica ion o Slo enian and non-Slo enian s awbe ies
was ob ained despi e di e en p oduc ion yea s using disc iminan
app oaches. Class models gene a ed by da a-d i en so independen
modeling o class analogy (DD-SIMCA) had high sensi i i y (96–97%)
and good speci ici y (81–91%) on a yea ly basis, while a mo e gene alized
model combining o al yea ly da a ga e a lowe speci ici y (63%) (24). O
he 33 comme cially a ailable samples ( es samples) wi h decla ed
Slo enian o igin, 39% we e om ou side o Slo enia. The speci ici y o
ga lic and aspa agus was ound o behighe compa ed o s awbe ies,
indica ing ha he model o hese wo commodi ies is mo e obus o
e i ying he co ec labeling.
These examples highligh he po en ial o iso opic and elemen al
analysis as eliable ools o ood o igin au hen ica ion while
demons a ing ha some commodi ies p esen mo e signi ican
challenges compa ed o o he s.
In addi ion, he IsoFoodT ack da abase can beused o suppo
ad anced analy ics based on s a is ical and explainable machine-lea ning
app oaches, enabling he de elopmen o disc iminan models o
di e en ia e selec ed ood commodi ies based on species using elemen al
and s able iso ope da a. Machine lea ning (ML) is a b anch o a i icial
in elligence (AI) ha enables sys ems o lea n and imp o e om
expe ience wi hou being explici ly p og ammed. The ML componen
accep s he ood’s iso opic composi ion as inpu and p edic s i s
geog aphical o igin. Addi ionally, his app oach o e s explana ions
abou which speci ic iso opes se e as indica o s o ha geog aphical
o igin, wi h he aim o inc easing us in AI-gene a ed sugges ions.
The me ada a included in he da abase allow he use o en ich and
complemen a s able iso ope e e ence da abase by a mo e no el
app oach, i.e., p ocess-based modeling, such as iso opic mapping
(isoscapes) (25). Isoscapes can be cons uc ed o make p edic i e
pa e ns and in o m he likelihood o o igin based on egional and
localized cha ac e is ics. The basic concep o isoscapes is e lec ed in i s
name, de i ed om he wo ds “iso ope” and “landscape.” Isoscapes
isualize he dis ibu ion o iso opic a ios ( ypically o ligh elemen s) in
space, o en using Geog aphic In o ma ion Sys em (GIS) echnology o
inco po a e hese a ios in o geog aphic maps. The e a e wo p ima y
me hods o p oducing isoscapes: s a is ical and p ocess modeling. In
s a is ical modeling, a ious geos a is ical app oaches, such as in e se
dis ance weigh ing and k iging, a e used o model he expec ed iso opic
composi ion o he ma e ial in ques ion. These me hods ypically equi e
ex ensi e da abases ha densely co e he a ea o in e es . Only a ew
na ional-scale isoscapes s udies such as wine (26), milk (27), oli e oil
(28), and ice (29) ha e been published.
Con e sely, p ocess modeling in ol es ob aining a iables wi h
highe spa ial densi y, such as me eo ological o geological da a, o model
he iso opic composi ion based on he p ocesses a ec ing iso opic
ac iona ion. Fo example, ood isoscapes a e de i ed om he
obse a ion ha ood p oduced in a speci ic a ea o en e lec s he local
clima ic and geological cha ac e is ics. The ad an age o p ocess-based
modeling o e s a is ical modeling is ha i equi es a much smalle
sample da abase and can beapplied o a eas wi h limi ed sampling. A
good example o he spa ial a iabili y “GIS modeling Isoscapes” o
oxygen and ca bon s able iso ope composi ion (δ13C, δ18O) o a gan oil
is also p esen ed by Taous e al. (30). In o de o make global, spa ially
con inuous p edic ions o a gan oil s able iso ope a ios, he mechanis ic
models in A cGIS so wa e (ESRI Co po a ion A cGIS 10.5) we e
implemen ed. The o dina y poin k iging was used o spa ially
in e pola e δ13C and δ18O alues o a gan oils om 25 indi idual samples
collec ed a i e independen egions.
These geospa ial models– isoscapes may p o ide a cos -e ec i e
ex ension o he iso opic da ase app oach.