Title: An Application of Serially Balanced Designs for the Study of Taste Samples with the aASTREE Electron
1An Application of Serially Balanced Designs for
the Study of Taste Samples with the a-ASTREE
Electronic Tongue
Stan Altan, Marc Francois, Sabine Inghelbrecht,
Areti Manola, Yan Shen Raritan, NJ, USA
Beerse, Belgium
Correspondence Email Yshen10_at_prdus.jnj.com
2Outline
- Introduction
- E-Tongue Experiment
- DOE- Serially Balanced Design
- Statistical Analysis
- Results and Discussion
- References
3Introduction
- Taste Perception is a biological process, where
taste buds are stimulated to send messages
through nerves to the brain. - Five basic tastes sweet, sour, bitter, salty and
unami (glutamate). - The pharmaceutical industry is most concerned
with bitterness, since a drugs bitter taste may
prevent patients from completing a therapeutic
course of treatment.
4Introduction
- Taste Testing Tools
- Human sensory testing.
- The drawbacks are
- Subjectivity of panel members.
- Time consuming, expensive.
- Potential toxicity of drugs.
- ?-Astree Electronic-Tongue (E-Tongue)
- Developed by French company AMOS during the past
six years, started to emerge within the past
three years in the pharmaceutical industry. - Multi-sensor system for liquids.
- Relies on chemical sensors and pattern
recognition.
5Introduction
- E-Tongue Mechanics
- An array of 7 electrochemical sensors and 1
reference electrode. - A liquid auto-sampler with a 16-position
carrousel. - An electronic unit for auto-sampler control.
- Software.
6Introduction
- How E-Tongue works
- Each sensor has a different organic coating,
which reacts with the test compound and generates
different potentials relative to the reference
electrode (Ag/AgCl). - The potentiometric difference between sensors
produces the measurement output. - Quantification is based on the combined response
across sensors thought to be related to the test
compounds chemical properties.
7Introduction
- Statistical Design Considerations
- Samples are presented in a serial order.
- Possibility of residual carryover effects of one
sample to subsequent samples, even with
intervening wash samples since the wash vessels
can become increasingly more contaminated by
repeated washings. - Ignoring carryover effects could bias the results.
8E-Tongue Experiment
- Description of the Experiment
- Four Test Samples and 4 Standard Samples.
- Eight Washes (designated W1, W2,,W8).
- Each Sample must be followed by a Wash.
- Sensors designated AB, BB, ZZ, JE, DA, BA, CA.
- Objectives
- Estimate residual and main effects.
- Relate Test Samples to Standard Samples.
9E-Tongue Experiment
- 4 Aqueous Test Samples
- C0 Water alone,
- C1 Sodium Topiramate 0.5 mg/ml (low bitterness),
- C2 Sodium Topiramate 5 mg/ml (middle
bitterness), - C3 Sodium Topiramate 10 mg/ml (high bitterness),
- 4 Aqueous Standard Samples
- S1 0.01 mM HCl (known acidic taste),
- S2 0.01 mM NaCl (known salty taste),
- S3 0.01 mM Sodium-L-Glutamate (known unami
taste), - S4 1 mM Quinine Hydrochloride (known bitter
taste).
10DOE-Serially Balanced Design
- Two types of serially balanced sequences,
referred to as Type 1 and 2. Type 1 includes
residual effect of the same treatment on itself,
Type 2 does not. - Can be derived from a Latin Square (Williams
Design). - Balanced for residual effect of minimal length.
11DOE-Serially Balanced Design
- Construction
- Start with a Latin Square with v2m treatments.
- Use the ith and (mi)th column (i0,1,,m-1)
omitting the 0th row, start from (2i1)th row,
forming a chain as indicated in the following
diagram. - Augment 0 to the beginning of the chain which
gives a Type 2 sequence. - Repeat each treatment once to obtain Type 1
sequence.
12(No Transcript)
13DOE-Serially Balanced Design
- Construction (Continued)
- If v is odd, v-12m is even and we construct a
balanced sequence in symbols 0, 1,,v-2 as before
and insert the symbol v-1 between 0 and 1, 1 and
2, 2 and 3, v-2 and 0. Then augment this added
sequence on the right by the symbols 1, 2,, v-2,
0.
14DOE-Serially Balanced Design
- Example for v8
- Start with a Williams Latin Square with v8
treatments
15DOE-Serially Balanced Design
- Example for v8
- Augment 0 to the beginning of the chain to obtain
a Type 2 sequence - 0 17263545362710 37465647302120
- 57675041323140 70615243425160
- Repeat each treatment once to obtain the Type 1
sequence - 0 1726354553627110 3746564733002120
- 5767750413223140 7066152434425160
16DOE for E-Tongue Experiment
- Use a Type 1 Serially Balanced Sequence to define
ordering of Samples. - Washes inserted into sequence of Samples
according to the row-column designation of an 8x8
Latin square. - Final design consists of a chain of Samples and
Washes of length 1281. - Each Sample and Wash evaluated 8 times with C0
evaluated 9 times to balance residual effects.
17Statistical Analysis
- Statistical Model Eestimates (by Sensor)
- Direct effects of samples and washes.
- 1st and 2nd order residual effects of samples.
- 1nd order residual effects of washes.
- Principal Components Analysis (PCA)
- Least squares means (lsm) from the model for each
Sample and Wash calculated by sensor, forming a
16?7 matrix. - The matrix of lsms used to calculate a 7?7
covariance matrix corresponding to the sensors in
preparation for a principal component analysis.
18Statistical Analysis
- Statistical model used to estimate direct and
residual effects
19Statistical Analysis
- Statistical Model (Continued)
20Figure 1 Pairwise Scatter Plot of 7 Sensors
21Figure 2 Serial Measurements of Samples by Sensor
22Results and Discussion
- Small or no residual effects were attributable to
Washes on Samples, - Residual effects of Samples on Washes were
frequently found to be statistically significant,
- Second order residual effects were found for
Sensors BB and BA, only with respect to Sample S1
(0.01mM HCL), - Direct Treatment effects compared to C0 for
Samples indicated frequent differences for
sensors CA, ZZ and BB, some difference for JE,
and none for BA, DA and AB.
23RESULTS Least Squares Means by Sample and
Sensor are given in the following table along
with results of significance testing of Samples
(compared to C0) and Washes (compared to each
other).
24Results and Discussion
- PCA The first component explained 87 of the
total variance and the second 12.
Figure3 Samples and Washes in relation to the
first two principal components
25Results and Discussion
- Washes were clustered together around the origin
as expected. - Control (C0) was very close to the washes since
they were basically the same. All other samples
were displaced from the washes. - Standard Samples were distinguishable with
respect to relative location but S4 (standard
bitter taste) was in the opposite quadrant to C1,
C2, and C3 (Topiramate). - Bitterness as a sensory quality arising from
different chemical compounds can fall in
different locations on the 2-dimensional scale. - PCA could not distinguish between C1 (low
concentration) and C2 (middle concentration) but
C1 and C2 were separated from C3 (high
concentration).
26References
- S. Altan, A. Manola, R. Pandey, J. Troisis, D.
Ragahavarao, A statistical design consideration
in robotic systems, Drug Information Journal
(2004), 3, 283-287. - S. Altan, D. Ragahavarao (2005) Serially Balanced
Designs for Two Sets of Treatments. Journal of
Biopharmaceutical Statistics 15(2) 279-282 - T. Uchida, A. Tanigake, Y. Miyanaga, K.
Matsuyama, M. Kunitomo, Y. Kobayashi, H.
Ikezaki, A. Taniguchi, Evaluation of the
bitterness of antibiotics using a taste sensor,
Journal of Pharmacy and Pharmacology (2003), 55,
1479-1485. - L. Zhu, R.A. Seburg, K. Thompson, E. Tsai, S.
Isz, Feasibility Study of an Electronic-Tongue
for Potential Pharmaceutical Applications - L. Zhu, R.A. Seburg, E. Tsai, S. Puech, J.
Mifsud, (2003) Flavor Analysis in a
Pharmaceutical Oral Solution Formulation Using an
Electronic-nose. Journal of Pharmaceutical and
Biomedical Analysis 34 453-461