APPLICATION OF CHEMOMETRICS FOR DATA PROCESSING OF THE ELECTRONIC TONGUE

1
APPLICATION OF CHEMOMETRICS FOR DATA PROCESSING
OF THE ELECTRONIC TONGUE

• Alisa Rudnitskaya, Andrey Legin, Kirill Legin,
  Andrey Ipatov, Yuri Vlasov
• Laboratory of Chemical Sensors, Chemistry
  Department, St. Petersburg University, St.
  Petersburg, Russia
• http//www.electronictongue.com

2
ELECTRONIC TONGUE RESEARCH GROUP
CHEMISTRY FACULTY RADIOCHEMISTRY
DEPARTMENT LABORATORY OF CHEMICAL SENSORS Head of
the Laboratory prof. Yuri Vlasov

• Project leader Dr. Andrey Legin
• Permanent Dr. Alisa Rudnitskaya
• staff Dr. Andrey Ipatov
• M.Sc. Boris Seleznev
• Associated researchers, currently 3 Ph.D.
  students, several students a year
•  

3
Research directions
1. New sensing materials
Solid-state materials (chalcogenide
glasses) Organic polymers Thin films
Electrochemical characteristics Cross-sensitivity
study Sensing mechanism
2. Chemical sensors
Multisensor arrays Chemometrics tools
Recognition Analysis
3. Sensor systems electronic tongue
4. Application of chemical sensors and sensor
systems
Industrial analysis Environmental control Medical
analysis Foodstuff analysis
4
Advantages and drawbacks of potentiometric
chemical sensors

•      Advantages
•  
• 1.  A wide range of available sensing materials
  and sensors.
• 2.  Wide variations of sensor properties, some
  unique features.
• 3.  A wide knowledge about composition/properties
  relationship.
• 4.  Simple installation. Easy, direct
  measurements.
• 5.  Different configuration (static, flow) and
  size (bulk, micro).
• 6.  Easy applicability for automatic routine
  analysis.
• 7.  Low cost.
•  
•      Drawbacks
•  
• 1.  Insufficient selectivity of many sensors.
• 2. The number of available sensors is far smaller
  than the variety of analytes.

5
Electronic tongue

• Electronic tongue is an analytical instrument
  comprising an array of non-specific, poorly
  selective, chemical sensors with partial
  specificity (cross-sensitivity) to different
  components in solution, and an appropriate
  chemometrics tool (method of pattern recognition
  and/or multivariate calibration) for the data
  processing. Of primary importance is stability of
  sensor behaviour and enhanced cross-sensitivity,
  which is understood as reproducible response of a
  sensor to as many species as possible. If
  properly configured and trained (calibrated), the
  electronic tongue" is capable to recognise
  quantitative and qualitative composition of
  multicomponent solutions of different nature.

6
Potentiometric electronic tongue
7
Electronic tongue laboratory version
8
Composition of chemical sensor array for
electronic tongue

• Chalcogenide glass sensors
• As2S3, GeS2, AsSe with various additives
• Polymer based
• PVC, plastisizer and active substances
• Chrystalline based
• Ag2S with different additives, LaF3
• Totally up to 40 sensors

9
Methods for the ET data processing

• Quantitative analysis (concentrations/parameters
  prediction)
• Modeling using MLR, PLS-regression, artificial
  neural networks, N-PLS
• Data exploration, recognition
• PCA
• Classification
• SIMCA, LDA, PLS-regression

10
Electronic tongue applications
Types of analysis Classification and
discrimination (identification,
recognition) Quantitative analysis of multiple
components simultaneously Process control Taste
assessment and correlation with human perception
Objects Food - fruit juices, coffee, soft
drinks, milk, mineral water, wine, vodka,
cognac, meat, fish, onion Medical analysis -
dialyses solution for artificial kidney,
pharmaceuticals, urine Environmental -
groundwater, seawater, dirty water from
farms Industrial analysis - galvanic baths, waste
purification systems, control of
biotechnology processes
11
Selected applications of the electronic tongue

• Discrimination of substances eliciting different
  taste and different substances eliciting the
  same taste
• Determination of ultra low activity of transition
  metals in seawater
• Determination of ammonium and organic acids
  content in the model growth media
• New approach to the data for flow-injection
  electronic tongue - determination of zinc and
  lead concentration in mixed solutions

12
Discrimination of taste substances

• Objective
• Discrimination of substances eliciting different
  tastes (i.e. bitter, sweet and salty) and
  substances eliciting the same taste
• Samples 10mmolL-1 individual solutions of
  substances
• bitter quinine, caffeine, drugs A and B
• sweet acesulfam K, aspartame, sucrose
• salty sodium chloride, sodium benzoate, drug D
• Measurements
• ET comprising 20 sensors
• at least 3 replicas of each sample in random
  order
• Data processing
• discrimination
• LDA
• PCA

13
Discrimination of taste substances
14
Determination of ultra low activities of
transition metals

• Objective
• Determination of ultra low activities of
  transition metals in waste waters and seawater
• Solutions
• Individual and mixed binary buffered solutions of
  Cu, Zn, Cd and Pb
• Total concentration of metals 1 ?M to 0.3mM,
  activity - 1nM to 0.1?M
• Background of 0.01M of NaCl and 0.01M citrate, pH
  8
• Measurements
• ET comprising 8 sensors
• Data processing
• Calibration and activity prediction of transition
  metals
• PLS-regression

15
Determination of ultra low activities of
transition metalsMeasurements in individual
buffered solutions
16
Determination of ultra low activities of
transition metals
17
Determination of ammonium and organic acids
content in the model growth media

• Objective
• Quantification of main substances consumed /
  produced during microorganisms growth
  monitoring of the fermentation processes
• Samples
• Set of 22 solutions modeling growth media
• Components MgSO4, KCl, KH2PO4, citrate,
  pyruvate, oxalate, glucose, glycerol, mannitol,
  erythritol, NH4Cl
• Measurements
• ET comprising 8 sensors
• At least 3 replicas of each solution
• Data processing
• Calibration and concentration prediction w.r.t.
  ammonium, oxalate and citrate
• Artificial neural network

18
Determination of ammonium and organic acids in
the growth media
19
Determination of zinc and lead concentrations in
mixed solutions using flow-injection electronic
tongue

• Objectives
• Evaluate relevance of different types of signals
  produced using flow-injection ET
• Evaluate relevance of different multivariate
  calibration methods for processing of the
  flow-injection electronic tongue data

20
Schematic of flow-injection electronic tongue
KNO3 0,1M
\
21
Flow-through cell
22
Sensor response parameters in FIA
ta- time before sample enters measuring cell ?
time of sample pass through the cell tb peak
width ?t- recovery time ? peak height
23
Data produced by flow-injection ET

• 1. Peak height measured for each sensor
• one signal from each sensor, I x J
• 2. Time-dependent response for each sensor
• Unfolded data set, I x JK
• 3. Time-dependent response for each sensor
• 3-dimensional data set, I x J x K

Time
Sensors
Samples
24
Calibration methods

• Data sets 1 and 2
• Partial least square regression
• Artificial neural network (back-propagation
  neural network)
• Data set 3
• N-way partial least square regression

25
N-PLS regression

• PLS-regression X TP E Y TQ E
• N-PLS regression X TWj(Wk) E Y TQ E

26
Experimental set-up

• ET 7 sensors with PVC plasticized membranes
• Set of mixed solutions containing zinc and lead
• Background solution - 0.1M KNO3
• Sensor potentials measured every 4 s for 2
  minutes, 30 points for each solutions
• Four replicas of each solutions
• Three types of data sets
• Data processing using PLS-1 and N-PLS regression

27
Sensors response in the individual solutions of
zinc and lead
28
Determination of zinc and lead in individual
solutions using flow-injection ET

• Calibration was done using PLS regression with
  test set validation, only pick height being used
  as sensor signals.
• Concentration range of both zinc and lead 10-6
  10-3 molL-1

29
Sensors response in the mixed solutions of zinc
and lead
30
Results of zinc and lead concentrations
prediction using three different types of data
sets
31
X-loadings weights
32
X-loadings weights
Time dependent response (3-d data)
33
Conclusions

• Use of time-dependent response of flow-injection
  ET instead of peak heights allows higher accuracy
  of concentrations determination in mixed
  solutions
• Use of 3-dimensional data set and N-PLS
  regression for calibration leads to simpler model
  and the same prediction errors compared to
  unfolded 2-dimensional data set and PLS
  regression for calibration