Development of NIRS Method for Predicting Starch Content in Sorghum Grains

Abstract

The potential for using Near-Infrared Spectroscopy (NIRS) as a way of quantifying starch in Sorghum grains was investigated. Spectral data from 41 ground Sorghum grain samples was obtained with an Antaris II FT-NIR analyzer (Thermo Scientific Inc., Madison, WI, USA) in reflectance mode. Each sample was averaged over 32 scans at a resolution of 4 cm-1 in the wavelength range of 4,000-10,000cm-1 (1,000-2,500nm.). The starch content for these samples was obtained through a Megazyme total starch assay kit, with two replicates for each sample. Of these 41 samples, 9 were randomly selected for validation, with the other 32 being used for calibration. Chemometric analysis was conducted by TQ Analyst 8.6.12 software (Thermo Scientific Inc., Madison, WI). The method of Partial Least Squares was used for model development. Model performance was evaluated in terms of the coefficient of determination (R2), root mean square error of calibration (RMSEC) and root mean square error of prediction (RMSEP). The Chauvenet test was used to eliminate outliers defined as the points at distances in the principle component space greater than 3.0. After eliminating outlier data, it was found that the calibration model has an R-value of 0.9224, translating to an R2-value of 0.8508, meaning there is a strong correlation between starch content and NIR absorptions. These results show that the NIRS model is promising for starch content analysis and can be used for the biofuel industry in the future.

Introduction

Many growing concerns such as petroleum price instability and dependence on imported oil have encouraged Renewable Energy acts to be put into motion by both the European Union and the United States. The European Union Renewable Energy Directive from 2009 set the following goals to be reached by the year 2020: having 20% renewable energy of total energy consumption, and a 10% fraction of renewable energy (ethanol) in the transport sector (Council of the European Union, 2009). The United States also adopted a renewable energy policy through the Renewable Fuel Standard (RFS) program. The RFS requires that renewable fuel be blended into transportation fuel in increasing amounts each year, escalating to 36 billion gallons by 2022 (Environmental Protection Agency, 2007). In order to meet these requirements potential energy crops such as wheat, rye, barley, and sorghum are being integrated into the current ethanol production process.

Sorghum is a promising energy crop due to its ability to tolerate drought, soil toxicities, a wide range of temperatures, and high altitudes (Stroade & Boland, 2013). In addition, it has short growing seasons, high

Sorghum Crop

Figure 1. Sorghum Crop

tolerance to pests and disease, and a grain yield equal to or even higher than that of other cereal grains (Rooney, Blumenthal, Bean, & Mullet, 2007). Over the past years, sorghum has been grown increasingly in the United States, with a total production of 432.6 million bushels in 2014 (United States Department of Agriculture, January 2015), meaning there is a plentiful supply of this grain to be used for ethanol production. A simplified model of this sorghum to ethanol process can be seen in Figure 2.

SorghumtoEthanol

Figure 2. Simplified Plant to Ethanol Process

The grain is harvested from the plant and ground into a fine flour. This flour contains a high proportion of starches, which are a subset of complex carbohydrates, or polysaccharides. Enzymes and yeast are then added to begin the process of fermentation. The enzymes break down the complex carbohydrates into smaller sugars that are able to be eaten by the yeast and converted into ethanol. After fermentation is complete, there will be a mixture of ethanol and leftover biomass called mash. At this point the mixture needs to be put through distillation in order to separate the ethanol from the mash. This process can take several days, and many different aspects need to be monitored to ensure everything runs smoothly. Since the amount of ethanol produced is based on the initial starch content, starch is a very important component to examine. Unfortunately, current laboratory procedures for starch quantification are time consuming and arduous, making them impractical for industry use . Fortunately, however, research investigating several components of various other cereal grains using Near-Infrared Spectroscopy (NIRS) has shown promise that NIRS is an excellent tool for quick component analysis.

Near-infrared spectroscopy (NIRS) is a spectroscopic method that uses the near-infrared region of the electromagnetic spectrum (from about 800 nm to 2500 nm) and is widely applied

LightSpectrum

Figure 3. Light Spectrum

in agriculture for determining the quality of forages, grains, and grain products, oilseeds, coffee, tea, spices, fruits, vegetables, sugarcane, beverages, fats, and oils, dairy products, eggs, meat, and other agricultural products. It is commonly used to quantify the composition of agricultural products because it meets the criteria of being accurate, reliable, rapid, non-destructive, and inexpensive (Burns & Ciurczak, 2007).

Many other studies have done research on cereal grains using NIRS. Kim and Williams were successful in proving NIRS as accurate when determining starch content and energy in canadian wheat, barley, and corn (Kim & Williams, 1990). Pohl and Senn investigated the potential of near-infrared spectroscopy for determining fermentable substance and ethanol yield in wheat, rye, and triticale grains and concluded that NIRS is an appropriate and useful tool for these predictions (Pohl & Senn, 2011). A study was also on whether NIRS can be used to predict the ethanol yield of dry-grind maize, and it was proven that a NIRS model could in fact be used to predict maize grain ethanol yield potential (Hao, Thelen, & Gao, 2012). Considering several other studies have had success in using NIRS as a method for cereal grain analysis, it is thought a similar model can be produced for sorghum grains. Therefore, the main objective of this project is to develop a NIRS model that can be used as a fast method for predicting the starch content in sorghum grains for further research and industry use.

Experimental Method

Several varieties of sorghum grain samples were received from a supplier in Texas. 41 of these grain samples were then scanned by an Antaris II FT-NIR analyzer

NIR

Figure 1. NIR Analyzer

(Thermo Scientific Inc., Madison, WI, USA) in reflectance mode. A sample cup spinner (Thermo Scientific, WI, USA) with an Integrating Sphere module was used to quickly and reliably obtain bulk information from the grain samples. Spectra were collected by rotating the sample cup through the NIR beam.

NIRinner

Figure 2. Inner workings of the NIR Analyzer

Each spectrum was averaged with 32 scans at a resolution of 4 cm-1 in the wavelength range of 4,00010,000cm-1.

Grain samples were then ground down to a 0.5mm diameter particle size using a UDY Cyclone Mill.

Mill

Figure 3. UDY Cyclone Mill

It was made sure that the mill, filter, screen, and collection jar were thoroughly cleaned after each run to ensure that there was no cross contamination between samples.

In order to calculate starch content, the moisture content of each sample must be known. Therefore, the moisture content of each sample was determined after milling. Two replicates were weighed out for each sample and placed in sterilized pans. These pans were then placed in a 130C oven for 3 hours. After 3 hours the pans were taken out of the oven and placed in desiccators to cool for 20 minutes. Once cool, the pans were weighed again and the following formula was used to calculate the moisture content for each sample: MC = (Wet Weight Total-Dry Weight Total)/(Wet Weight Flour)X100%

Oven

Figure 4. Oven full of samples being tested for moisture content

The next procedure to be carried out was starch content analysis. A total starch assay kit from Megazyme was used to quantify the starch for each sample, with two replicates per sample. The main enzymes used were alpha-amylase and amyloglucosidase. The final part of the procedure requires that the samples be run through a spectrophotometer, which measured the absorbencies at a wavelength of 510nm. These absorbencies were then entered into a Mega-Calc spreadsheet along with the samples weight and moisture content. The spreadsheet then returns the starch content as a percentage of the dry weight base for the sample.

Once all of the data was collected, a chemometric analysis was conducted using TQ Analyst 8.6.12 software (Thermo Scientific Inc., Madison, WI). The pathlength was calibrated using standard normal variate (SNV) and multiplicative signal correction (MSC). The Savitzky-Golay filter was used to reduce random noise. First and second derivatives were used as a pretreatment method to resolve spectra peak overlap and eliminate linear baseline drift. The first derivative was the rate of change of absorbance with respect to wavelength, whereas the second derivative corresponded to the curvature or concavity of the graph. First and second spectra formats were compared. To avoid bias in the subset and achieve a calibration set and validation set with a ratio of 3, all 41 sorghum grain samples were sorted ascendingly by measured value. One in every five samples was randomly assigned for validation (9 total) , with remaining samples as calibration samples (32). A full spectra range from 4,000 to 10,000 cm-1 was used unless otherwise specified. The method of partial least squares (PLS) was used for model development. Model performance was evaluated in terms of the coefficient of determination (R2), root mean square error of calibration (RMSEC), root mean square error of prediction (RMSEP), and ratio of standard deviation of calculated set (SDy) to RMSEP (RPD).The Chauvenet test was used to eliminate outliers defined as the points at distances in the principle component space greater than 3.0. Predicted residual error sum of squares (PRESS) diagnostic function was used to determine the number of factors necessary for calibration.

Results and Discussion

Spectrum

Figure 1. Grain Spectra

FSpectrum

Figure 2. Flour Spectra

It is noticeable that these spectra have a very similar general shape to them, however it is also quite obvious that the grain spectra are much more spread out than the flour spectra. This could be due to the fact NIRS is more accurate when reading spectral data from a more uniform sample such a flour as opposed to grain. It can also be seen where the peaks aren't as smooth that there is some noise in the background of both data sets. This noise can be cleaned up by processing more samples and re-processing some of the outliers.

MC

Figure 3. Moisture Content of flour samples

Spectrum

Figure 4. Starch Content of flour samples

The average moisture content for sorghum grain flour is 8-12%, and should not exceed 15% (Dicko, et al., 2006). The starch content of sorghum ranges from 56-75% (Jambunathan & Subramanian, 1988). Therefore, it is safe to say that the data acquired is appropriate to use in the development of a NIRS model for starch content prediction.

GM

Figure 5. Grain Calibration Model

FM

Figure 6. Flour Calibration Model

The calibration model for the sorghum grain samples has an R-value of 0.7829 translating to an R2 value of 0.6129. In addition, the calibration model for the sorghum grain flour samples has an R-value of 0.9224, meaning it has an R2 value of 0.8508. This information makes it apparent that the model for sorghum grain is not as accurate as the sorghum flour model. However, since the R2 value for the sorghum grain flour model was 0.8508, it shows that there is a strong correlation between starch content and NIR absorptions.

Conclusions

There was hope that a NIRS model could be derived for sorghum grain as a non-destructive method for starch analysis, however the data shows that the samples may not be uniform enough to acquire an accurate reading. On the other hand, even though there is a noticeable amount of background noise in the NIR spectra, the data from the calibration model for sorghum grain flour shows a strong correlation between starch content and NIR absorptions. These results demonstrate that the NIRS model for sorghum grain flour is promising for starch content analysis and can be used for the biofuel industry in the future.

For future research, more sorghum grain and flour samples will be analyzed and added to the starch calibration models. The additional samples will increase the accuracy of the models and decrease the background noise . In addition, we believe the R2 value will increase as the number of sorghum samples increases.

About Me

  • University of Kentucky
  • Biosystems Engineering, Junior, 2017
  • Flaherty, Kentucky
  • Outdoor sports/ activities, cooking, baking

FM

Me and one of my dogs

Acknowlegements

  • Kaelin Saul, IGERT Fellow and Graduate Student, Department of Biological and Agricultural Engineering, Kansas State University
  • Dr. Donghai Wang, Professor, Department of Biological and Agricultural Engineering, Kansas State University
  • This material is based upon work supported by National Science Foundation Grant: REU Site: Summer Academy in Sustainable Bioenergy; NSF Award No.: SMA-1359082.

References

Refs