A Brief Introduction to NIR Spectroscopy
Spectroscopy) is the technique of using a sample's NIR absorbance characteristics to predict parameters of interest.
NIR is a region of the Electromagnetic spectrum. Click the chart below to enlarge the graphic.
NIRS exploits the fact that many natural products absorb NIR radiation at specific regions or wavelengths. Specifically, N-H, O-H and
bonds are strongly absorbed by NIR radiation, with other molecular bonds less so. Some known absorption bands in the NIR are shown in the chart below.
Thus, samples high in proteins (many N-H bonds) will absorb more in the amine (N-H) bond regions than samples low in protein. Samples high in moisture and or / sugars will have higher adsorptions in regions associated with hydroxyl (OH) bonds. A sample's NIR spectrum will be a composite of all the absorbances from all of the molecular bonds in the sample. A simplified diagram is below.
This is a very simplistic explanation for how NIRS works. With the thousands of different molecules in the average natural product, the information in the NIR spectra is very complicated and difficult to interpret. Routine NIRS analysis would not be possible without computers and chemometrics. Spectra from NIRS instrumentation, chemometrics, and reference values are used to calibrate NIRS instruments.
To create a calibration, samples are chosen which are representative to the samples that will be analyzed. There may be as few as 60 samples or there could be several thousand. These samples are analyzed in the NIR, and the spectra stored. The samples are then sent to have the reference analysis performed on them. These are termed calibration samples. We will use 60 sample analyzed for protein content and scanned on a NIRSystems model 5000 for this example.
When the samples come back, the spectroscopist has:
- 60 sample spectra consisting of 700 datapoints (1100 - 2500 nm)
- 60 sample reference data consisting of protein content values
To create a calibration, a mathematical relationship can be established between these two sets of data. This can be done via several chemometric techniques.
||Multiple Linear Regression
||Partial Least Squares
||Artificial Neural Network
Each of these chemometric techniques establishes a mathematical relationship between variation in the NIR spectra of samples with the variation in the parameter measured. This relationship can then be used to predict the parameter value in unknown samples.
Some key points about NIRS:
- It is a secondary technique that is calibrated against a primary reference method. Usually this is a wet laboratory assay.
- It is very good at predicting samples that are similar to those that were used in the calibration. Conversely, samples very different from the calibration samples many times are not predicted reliably.
- The use of outlier identification is important to ensure that the unknown samples belong to the calibration you are using to analyze.
- NIRS is not magical nor mysterious, but depends on solid mathematical relationships between the NIR spectra and the parameter of interest.