Near Infrared reflectance spectroscopy (NIRS) Dr A T Adesogan Department of Animal Sciences University of Florida
Benefits of NIRS Accurate Rapid Automatic Non-destructive No reagents required Suitable for large nos of samples Characterizes the entire forage composition
Prediction of in vivo OMD of 122 silages from different methods Method r 2 RSD (M) ADF 0.34 0.051 Pepsin + cellulase 0.55 0.042 ISD (48hr) 0.68 0.036 Rumen fluid-pepsin 0.74 0.032 NIRS 0.85 0.024 (Barber et al., 1990)
Indices successfully predicted by NIRS Lactate & VFA content N degradability Soluble N & NH 3 N Feed intake, digestibility and ME content Minerals Oil and CP content ADF content Lignin content Lignin composition Alkaloids Fungal contaminants Fermentation characteristics GE content Botanical composition Effect of NH 3 treatment
Petersen 2002 (Foss Tecator)
Underlying principle Based on using wavelengths relating to the absorbance of light by chemical components within the feed to predict nutritive value Forage reflectance spectrum correlated against standard samples of known composition to derive a relationship that can be used for future predictions.
Principle (Williams, 1977 )
Absorbance NIR region = wavelength range 700-3000 nm Conventional NIR machines for forage evaluation use the 1100 2500 nm wavelength region NIR spectra are plots of reciprocal log 10 reflectance (log 1/R) versus the wavelength
NIR spectrum (Deaville and Flynn, 2000)
Wavelength choice Most important determinant of accuracy of predicting forage quality Based on understanding the wavelength regions associated with various chemical constituents Choice should: reflect constituents which are part /relate to the predicted term minimize the number of wavelengths
Prominent wavelength regions Water 1940 & 1450 nm Aliphatic C-H bands 2310, 1725, 1400nm Lipids O-H bands 2100 & 1600 nm Carbohydrates N-H bands 2180 & 2055 nm Proteins (Deaville & Flynn, 2000)
Developing the calibration Calibration = regression b/w spectra or wavelengths & predicted term (e.g. intake) Process 1. Examine population structure Must include all possible variation in future samples 2. Choose the relevant wavelengths 3. Employ a math treatment to develop the calibration 4. Validate the calibration
Math treatments Multiple linear regression Adds variables to a monovariate regression Possibility of overfitting/ math artefact predictions + Uses only limited spectral information Gives less accurate predictions Principal components analysis (/regression) Groups spectral data into a few, independent components which are used as the predictors Hence uses most of the spectral data More accurate Multiple partial least squares Similar to PCA but uses both lab data & spectral data in the prediction Often most accurate
Effect of math treatment on S.E.P Math treatment Voluntary DMI OMD MPLS 6.57 38 PCA 6.68 41 MLR 7.37 40 S.E.M 0.123 0.9 P <0.05 <0.06 (Deaville & Flynn, 2000)
Validation of the calibration Entails testing the calibration on a different data set Conventional method Uses an independent population for validation Requires large # of samples (preferably >100) Internal cross validation method Separates the population into different groups and Progressively develops calibrations b/w groups & reference data till validation is complete Copes with smaller sample sizes
Factors affecting NIRS results Wavelength choice Math treatment NIR instrument type Sample preparation (density, particle size, % moisture) Spectral data pretreatment techniques
NIR instrument types Scanning monochromators Scan the entire wavelength regions Measure at 700 spectral points= more accurate Fixed-filter instruments Cheaper hence favoured by some labs Measure at fewer spectral points Only accurate for predicting well-defined chemical entities hence of limited use for digestibility predictions Can overcome this by developing relationships b/w fixed-filter instruments & monochromators
Misleading predictions due to sample moisture % Using Wavelengths b/w 1450 and 1620 nm in calibration enhances prediction of hay digy (Coleman and Murray, 1993). However, water is also absorbed in the this region This highlights the need for proper elimination of moisture or use of undried samples.
Effect of milling on S.E.P Method DMI OMD Coarse milling 7.88 41 Finely milled, 5.97 37 S.E.M 0.349 1.4 P <0.001 <0.001 (Deaville & Flynn, 2000)
Spectral data pre-treatment Forages/ feeds give overlapped absorption bands rather than sharp individual peaks at specific wavelengths Spectral data pre-treatment can resolve such problems which are due to: Sample particle size variations Temperature/humidity Light scatter Path length variation
Light pathways (Reeves III, 2000)
Spectral shifts A B = Peak shifts Can be due to temp. variations A C Baseline shifts Can be due to particle size variations 1 2 3 (Reeves III, 2000) A D Multiplicative scatter Can be due to particle size variations A E Multiplicative scatter 2 nd component (F) present
Shift correction methods Correction Methods include: Derivatization Std. Normal variate detrending Multiplicative scatter correction
Derivatisation When NIR spectra contains several overlapped bands Derivatisation resolves overlapped bands into component absorptions Hence derivatisation increases peak definition Reduces the effect of variable path length
Derivatisation
Other spectral pre-treatments Standard normal variate (SNV) detrending Scales each spectrum to have a s.d. of 1.0 Reduces spectral & particle size variability Repeatability file/ multiplicative scatter correction Re-shapes each spectrum & till it resembles the target spectrum obtained from the mean of a file of spectra Reduces variability due to moisture content
SNV detrending Accentuates moisture content 0.7 effect 0.6 0.5 Log 1/R 0.4 2 0.3 0.2 1 0.1 1000 1200 1400 1600 1800 2000 2200 2400 2600 Wavelength (nm) SNV-D 0 Raw Spectra -1-2 1000 1200 1400 1600 1800 2000 2200 2400 2600 Wavelength (nm) SNV-detrended Spectra
NIRS - problems Expensive initial outlay Black box biological meaning Requires large data sets & frequent updating Transfer of wet chemistry errors Calibration population must be similar & contain same variation as samples to be tested.
NIRS - problems Species-specific equations Can t be directly used for predicting mineral % Minerals not absorbed in the NIR region Can only use NIRS for minerals based on correlation b/w the mineral and an organic component
NIRS problems continued Requires validation Most analytical methods also do, but this is ignored Complex algorithms/ chemometrics required Misuse of equations Species-specific equations used for other spp Calibrated with unvalidated reference methods
References Deaville and Flynn, 2000. Near infrared reflectance spectroscopy: An alternative approach to forage quality evaluation. In Givens et al. 2000. Forage evaluation in animal nutrition. Page 201. CABI, Wallingford Reeves III J. B. 2000. Use of near infrared reflectance spectroscopy. In D Mello JPF. Farm animal metabolism and nutrition. Page185. CABI Publishing.