Types of Data : Spectral & Non-Spectral
CAMO’s multivariate statistical software products, solutions and services contain comprehensive analysis for Exploratory Statistics, Regression Analysis, Classification, Prediction, and Design of Experiments.
With a defined focus on Multivariate analysis, our software products and solutions enable an intelligent analysis of -
- Spectral Data (Spectroscopic, Chromatographic)
- Non-Spectral Data (Sensory, Environment etc)
Spectral Data
Spectral Data is essentially data derived by the use of spectroscopic instruments like NIR, FTIR, UV-VIS and chromatographic instruments like HPLC and GC. This data specifies variables with its properties and undoubtedly provide a great deal of useful information about organic molecules.
Scientists use spectral data to:
- Discover the chemical composition of materials by looking at the light (and other kinds of electromagnetic radiation) they emit
- Identify and monitor the production of products in factories
Examples of Spectral Data
Flavour analysis relies heavily on gas chromatography and gas chromatography-mass spectrometry to separate and characterize flavour components. An effective way of finding out which flavors are released from a food is to carry out a static headspace analysis, which uses a gas chromatograph (GC) to measure how flavour compounds are split between an aqueous solution of volatile flavour compounds and the air above the solution (the headspace). The advent of GC machines with non-destructive detectors has made flavour research more interactive, and some GCs now incorporate sniffing ports to enable trained assessors to smell individual flavour compounds as they are separated by the GC.
Non-Spectral Data
Non-Spectral Data is essentially data collected from sensory and environment. Any data that is collected from other sources, than spectroscopic and chromatographic instruments is non spectral data.
Non-spectral interferences include physical and chemical interferences. Physical interferences are effects associated with the sample nebulization and transport processes. Changes in viscosity and surface tension can cause significant inaccuracies. Chemical interferences include molecular compound formation, ionization effects, and solute vaporization effects.
Scientists use non-spectral data to explore the physical characteristics and tangible qualities of products.
In the flavour industry, using the more traditional sensory panels of trained assessors, who can taste a food and decide which descriptors (spicy, rancid, fruity, green etc) best describe it, would be an example of non-spectral data.
Spectral and Non-spectral data play a crucial role in the following industry verticals:
- Pharmaceutical & Biotech
- Agriculture and Environmental
- Oil and Gas
- Food Science and Nutrition
- Chemicals
- Polymers
- Specialty Materials
Case Study
Assessing the Composition of Dairy Products and Grain by Near Infrared
Near infrared (NIR) spectroscopy can be used to replace wet chemistry in quantifying many compositional parameters of foods. Analysis of a sample using a traditional wet chemistry approach requires hours, but with the use of optical spectroscopy coupled with chemometrics data processing (using The Unscrambler 8.0® software), the time is cut to 60 seconds making the technique conducive to on-line or near-line process quality control. The data outlined in this note demonstrates the use of NIR to assess the quality of cheese and wheat samples, but the technique has been proven effective for prediction of fat, moisture, protein and fiber in a diverse array of food products such as soy beans, fish meal, and ground meats.
Using the NIR/PLS approach detailed in this application note, the test is non-destructive, quick, inexpensive, and there is very little difference in the results when different technicians are used.
Background
Wet chemistry for protein analysis is traditionally done by the Kjeldahl method and requires approximately 6 hours performing a test. Moisture evaluation requires between 1 hour and 72 hours depending on the nature of the sample (for example, moisture in the cheese data presented here required a 15 hour test). Fat is the most involved and technically demanding of the tests with a professional chemist averaging between 8 and 12 samples per day using the modified Babcock procedure (although there is a less-accurate instrumented fat measurement where the throughput is more like 20 samples per day). Because the quality analysis is a laboratory procedure, the sample throughput is constrained to a handful of analyses per day. When an upset occurs in the process line, the delay can often mean that there will be significant wasted product before the results are known and the process can be corrected. On the other hand, a typical NIR instrument requires 30 seconds to load and 30 seconds to measure and process the data, quickly alerting the operator to any process change. The basic issues to be addressed in this application note is whether the NIR instrument can match the accuracy of a wet chemical evaluation (as illustrated in Figure 1).
To test the effectiveness of near infrared results for evaluating moisture and fat content, a succession of brick cheese batches were analyzed by both instrumental and wet chemical techniques. In this example, 12 absorbances in the shortwave near infrared plus two process temperatures were measured for 140 random samples of the product. The resulting NIR spectral data was assembled into a file. Moisture and fat content (the latter by modified Babcock) were measured to supply reference values. A PLS model for moisture and fat content of the cheeses was created using the NIR results. The resulting comparisons of wet chemistry and instrument results are shown in Figure 2. The validity of the instrument model was tested (using a leave-one-out cross validation technique) and was found to be comparable to the wet chemistry assessment.
The near infrared instruments used in the analysis collected discrete absorbance values in the shortwave near infrared (918nm to 1050nm). Three of the four data sets were monitored with 12 wavelengths; one cheese data set was from a 7 wavelength system. The Unscrambler 8.0 software was used to model the data using the Partial Least Squares (PLS) and Principal Component Regression (PCR) techniques.
Analysis of Cheese
The quality of cheese is monitored by measuring the fat and moisture contained in the product. In essence, the fat content is a measure of the richness of flavor in the product, moisture relates to texture (and profitability). It is desirable to test as much of the product as possible in order to control the quality closely, but cost is the trade-off.
The NIR predictions were always within 0.5% of the measurements derived from drying and show a standard error that is somewhat better than the results for brick cheese at 0.23%.
Comparison of NIR predictions to the results of wet chemistry are much the same. However, the standard error of prediction for the analysis was 0.33% for the assessment of moisture and 0.31% for fat. In both the brick and cheddar examples of the fat content, as measured by the modified Babcock procedure, that yields values in 0.5% increments. This resolution makes the measured values appear to be discrete rather than continuous measures of fat content. The error value for fat should be considered as an upper limit for the technique inasmuch as the resolution on the wet chemistry is only 0.5%, thus affecting the ability of the NIR model to perform to a level better than the resolution allows. NIR/PLS Laboratory Brick Mozerella Cheddar Average Range.
The NIR technique can clearly identify this sample as out of standard range, but the prediction of fat content will be subject to increased errors unless additional data points are collected in this region to give a more detailed calibration.
The plots of predicted versus actual show a clear correlation between the modeled values of fat and moisture and the wet chemistry results. The standard error for the NIR results using PLS was computed to be 0.44% for the moisture results and 0.32% for fat. In comparison, errors in a model created by an alternate technique (PCR) were larger, displaying standard errors of 0.47% and 0.37% for moisture and fat, respectively. All values were within the expected variation for the wet chemistry of approximately 0.5% for both measurements within the range of values seen in the brick cheese data. To confirm the validity and robustness of a NIR/ PLS approach to assessing moisture and fat in cheese, two additional data sets were analyzed. One set contained the moisture results for 110 samples of mozzarella cheese (as assessed by 7-wavelengths in the NIR), the other 150 samples of cheddar cheese (using 12 wavelengths).
Analysis of Wheat
An evaluation similar to the cheese data was applied to wheat samples and NIR was used to evaluate the moisture and protein content of 135 samples. The NIR model is based on data that include 12 wavelengths and 2 temperatures. As with the analyses of the three types of cheese in the previous section, the purpose of this study is to see how closely the near infrared technique can match the precision of wet chemistry methods. The results were comparable with standard errors for moisture of 0.42% and for protein of 0.40%. Again, the results compare favorably with the laboratory method and fall within the guidelines set by the U.S.D.A.
Summary
We have demonstrated the effective use of a simple near infrared spectrometer to quantify the composition of wheat and cheese products. The technique requires that a model be built calibrating wet chemistry results to the instrument data using chemometric data processing techniques (in this case Partial Least Squares modeling). After the calibration step, the spectrometer can be used to augment or replace the wet chemical techniques for routine quality control. The standard errors of NIR instrumented assessment of moisture, fat and protein match those of the wet chemistry techniques. The advantage of the NIR approach is that it is requires less operator time, is more independent of the operator technique and is less expensive to run and maintain the equipment. In addition, the NIR approach does not require the same level of laboratory support and is conducive to on-line or near-line evaluation of a variety of food products.
