A lot
of organic dark chocolate were
characterized by a well perceivedmould
flavour. The aim of the study was to
characterize aroma componentsto improve
a control plane on supplier chain to
prevent the production of badflavoured
chocolate.The study started from edible
dark chocolate to continue, in the
future, tococoa bean.
Sampling
Some different lots of organic dark chocolate were sampled from
warehouse. There are three type of samples: defected samples (D), blank
samples without defects (B) and not known samples.
Sensory profile
(ISO 13299:03)
The samples were analyzed three time by a well trained panel of twelve
experts judge (Afnor V09A), without randomization plan.
The training were done by the use of specific standards developed by the
way.
The standards were: cocoa mass, well-nibs, mould-nibs, 70% chocolate
(crunch), cocoa powder without KOH treatments, 90% chocolate (bitter).
Objectives
The aim of the work is
to
estimate a chemometric
model
to prevent the
production of offflavoured
chocolate bar.
Methods
The sample of
chocolate (50g)
were stored at 25°C
for 30 min.
SPME fiber Carboxen/PDMS
Stable Flex 85μm (Supelco:
57334-U) previously
conditioned
at 300C
with carrier helio for 1
hour, was exposed at
headspace of the
chocolate
samples for about 30
min. at
25°C.
Injection with split/splitless
GLC
(6890N) MS (5973N)
Agilent
Condition GLC:
- column: Varian Factor
Four
VF-1ms 60m, 0,32mm ID,
1,0m
(CP8930)
- gas pressure carrier:
10 p.s.i.
- type of injection:
splitless for
0,5 min.
- injector temperature:
250°C
- oven temperature:
initial
temperature: 45°C
for 1 min
scale
5°C/min
to 250°C
for 1 min
Condition Mass Spectrometer
(wiley libraries):
- transfer line temperature:
280°C
- quadrupole temperature:
150°C
- source temperature: 230°C
- autotune optimization
- EM offset: 200 volts
- SCAN: 35-350, threshold
150,
sampling rate 2^2
Results
Sensory profile put in
evidence that only one good sample was
very good, the other two are affected by
very low intensity musty/mould defect.
In presence of off-flavour in the sample
is not possible to apply the
randomization plans to avoid the
carry-over effect, because the samples
must be divided into two groups: the
defect one and the no defect one. The
CV% are low and the panel work good in
the
three session, ANOVA
give a not significant response through
the session and repeatability is good
(IR<<3). The mean of the three session
was used to next computations. The radar
profile show the five sensory profile of
the analyzed samples.
On the other way, the
SPME/GC-MS analysis
gives
the probable marker
molecules
of the defect. It is
interesting to
denote that one of these
is very
linked to the presence/absence
of the defect (butenamide,
Noxopropyl)and others are
present in the
particular defect
samples (ethanol/phormic
ac.;
pentanal, 2-hexanal).
The only practical way
to
assemble chemical and
sensory
data with low cases (samples)
is
to use the PLS
regression
model.
PLS is a method based on
recursive computation
very
useful for rectangular
data
matrix with a serious
problem to
compute the inverse
matrix. It
derive from Yoreskog-H.
Wold
study of the seventh’s,
today it
has an incredible
evolution in
sensory analysis and
chemometrics with
studies of S.
Wold, Tenhenaus and
Esposito Vinzi.
Conclusion
The results of the
implementation of a PLS regression
model is the figures
above.
It is clear that the
marker of musty/mould is the cyclic
amide in association
with ethanol/phormic ac., pentanal
and 2-hexanal.
The prediction of the
model is very good, in fact the two
confidence bound on the
straight line are close one to
another, this shows a
good fit of the model.
The model must be
applied to nibs to predict the future
of the chocolate bar.
Presented at 8th
Pangborn Sensory Science Symposium
26-30 June; Florence, Italy.
