Characterization Of Different Unconventional Protein Feed Materials

Utilization of unconventional protein feed materials could be an effective way to alleviate the shortage of protein feed in China. However, rapid screening methods for characterization of dried distillers’ grains (DDG), dried distillers’grains with solubles (DDGS) and meat and bone meal (MBM) cannot fulfill the current needs. In this research, representative DDG, DDGS and MBM samples were collected and their characteristics of chemical compositions were studied. Chemometrics and near infrared spectroscopy/microscopy were used to characterize their properties. This work will provide a theoretical basis and technical support for further utilization of DDG, DDGS and different species of MBM. The main contents and results of this thesis are as follows:The chemical compositions characteristics of DDG samples were studied and rapid quantification models based on near infrared spectroscopy (NIRS) were established. Results have shown that the chemical compositions of DDG have broader range and great variation. NIRS models of crude fibre, crude ash, crude protein and different kinds of amino acids were accurate, the coefficient of determination in validation were greater than0.91, the relative prediction deviation (RPD) were greater than3. All the quantification NIRS models can be used in situ. This study provided rapid analysis method for characterizing DDG and it will benefit for utilization of DDG in the future.The use of genetic algorithm (GA) and backward variable selection partial least square (BVSPLS) methods to select the spectral variables and construct NIRS calibration models for determining protein content in corn DDGS were evaluated. GA and BVSPLS analysis selected8%and16%of the total NIR spectral variables respectively; the standard error of prediction (SEP), RPD of corresponding GA model and BVSPLS model were0.82%and0.66%,3.38and4.20, respectively. The performances of GA model were comparable to that of the full spectrum model, while the BVSPLS model significantly improved the accuracy of the model fit. By using variable selection method, the number of spectral variables used in the NIR calibration model were reduced and the interpretation of the correlation between the protein content of corn DDGS and selected spectra variables were improved. These results also have important implications for the development of a rapid online analysis system to detect protein content of corn DDGS in-situ.The feasibility of NIRS combined with chemometircs to classify geographical origin of corn DDGS was studied. Results have shown that partial least square discriminant analysis model established with standard normal variate combined with first derivative (15points smoothing) pretreated spectral data can perfectly discriminate samples from the Europe, the United States, the Jilin Province of China and the Heilongjiang Province of China. Combinded the results of chemical composition analysis with spectral variables that contributed most to the separation of samples from different origins, the reason of NIRS to classify DDGS from different geographical origins were preliminary interpreted. This study provided new method for tracing the geographical origin of corn DDGS. Self organizing feature maps (SOFM) was introduced to characterize amino acid composition of different species of processed animal proteins, including fish meal, poultry MBM, pig MBM and ruminant MBM (bovine and ovine). Results have shown that SOFM can effectively extract amino acid composition characteristics of different species of processed animal proteins and visually presented their similarities and differences. Poultry MBM, pig MBM and ruminant MBM have very similar amino acid compositions, while MBM and fish meal have very dissimilar amino acid compositions. Great attention should be paid to the balance of amino acid compositions when using meat and bone meal to replace fish meal in animal diets.Based on near infrared microscopy and consensus modeling, rapid method for identification species origin (poultry, pig and ruminant) of MBM sediment particles were established. The proposed consensus modeling strategy integrate those results obtained from K nearest neighbor model, partial least squares linear discriminant model, support vector machine discriminant analysis model and X-Y fused neural network model. It effectively unified inconsistent results from different discrimination models and improved the reliability of the prediction. The consensus modeling given out a99.40%prediction accuracy to the validation set of samples. This study have important implications for solving multi-classification problems.