The Use Of Artificial Neural Network For Modeling In Vitro Rumen Methane And Volatile Fatty Acid Production

The anthropogenic greenhouse gase emissions including methane (CH4) and carbon dioxide (CO2) not only cause atmaspheric greenhouse gas effect but also enhance global warming. Anaerobic enteric fermentations in ruminants are one of the principle sources of CH4and CO2from agriculture, which leading to animal energy intake loss and lower energy digestibility. Accurately predicting greenhouse gas from enteric fermentation in ruminants is the basis for evaluating greenhouse effect. In total,4experiments were conducted to develop different types of prediction models through analyzing the relationships between feed ingredients and the CH4, CO2, and volatile fatty acid (VFA) production in rumen fermentation. The performance of these models was tested. The suitability of using artificial neural network (ANN) to predict CH4, CO2, and VFA production in rumen fermentation were evaluated by comparing the tradictional regression equations and ANN models. Thus the prediction models with greater accuracy were proposed.Exp.1was conducted to develop2datasets, of which forty-five feed mixtures of the first dataset were formulated for beef cattle with concentrate/roughage ratios of10:90,20:80,30:70,40:60, and50:50using as feed samples for modeling dataset, and nine feed mixtures were formulated for each conentrate/roughage ratio. Ten feed mixtures of the second dataset with the same concentrate/roughage ratios with the dataset for modeling were formulated for testing the models, of which two feed mixtures were formulated for each conentrate/roughage ratio. The Menke and Steingass’s gas test was used for the measurement of CH4and VFA production. The feed samples were incubated for48h and the CH4, CO2, and VFA production were analyzed using gas chromatography.Exp.2The relationship between the CH4production and the Cornell Net Carbohydrate and Protein System (CNCPS) carbohydrate fractions of feeds for cattle and the suitability of CNCPS carbohydrate fractions as the dietary variables in modeling the CH4production in rumen fermentation were studied. Two datasets used were the same as Exp.1. The CH4production and CNCPS carbohydrate fractions of forty-five feed mixtures of the dataset for modeling were used for developing multiple linear equations and that CNCPS carbohydrate fractions often feed mixtures of the test dataset were used for validating the equations. Results indicated that the CH4production (mL) was closely correlated with the CNCPS carbohydrate fractions (g), i.e. CA (sugars), CB1(starch and pectin), and CB2(available cell wall) in a multiple linear pattern:CH4=(89.16±14.93) CA+(124.10±13.90) CB,+(30.58±11.72) CB2+(3.28±7.19), R2=0.81, P<0.0001, n=45. Validation of the model using ten feed mixtures indicated that the CH4production of the feed mixtures for cattle could accurately be predicted based on the CNCPS carbohydrate fractions. The trial indicated that the CNCPS carbohydrate fractions CA, CB1, and CB2were suitable dietary variables for predicting CH4production in rumen fermentation in vitro.Exp.3was conducted to investigate the suitability and accuracy of modeling the rumen CH4production of feed mixtures for cattle using three layer back propagation (BP) artificial neural network (ANN) which contained input laryer, hidden layer, and output layer. Two datasets used were the same as Exp.1. The CH4production and CNCPS carbohydrate fractions of forty-five feed mixtures of the dataset for modeling were used for training the ANN models and that CNCPS carbohydrate fractions of ten feed mixtures of the test dataset were used for validating the ANN models. The ANN architectures with1-16hidden layer nodes were compared and the effective ANN architectures were established. Paired t-test showed that no difference was found between the predicted and the observed CH4, CO2, and total gas production based on the ANN models (P>0.05). Model performance analysis based on the test data showed the root mean square prediction error (RMSPE) were3.89%,2.95%, and4.23%and the determination coefficient (R2) between the predicted and the observed values was0.95,0.97, and0.92for CH4, CO2, and total gas production, respectively. Validation of the ANN models indicated that the in vitro CH4, CO2, and total gas production of feed mixtures for cattle could reliably and accurately be predicted based on the CNCPS carbohydrate fractions using ANN models.Exp.4aimed to develop multiple linear regression (MLR) models and three-layer Levenberg-Marquardt back propagation (BP3) neural network models using the CNCPS carbohydrate fractions as dietary variables for predicting in vitro rumen VFA production and further compare MLR and BP3models. Two datasets used were the same as Exp.1, of which the dataset for modeling containing VFA production and CNCPS carbohydrate fractions of forty-five feed mixtures with concentrate/roughage ratios of10:90,20:80,30:70,40:60, and50:50were used for establishing the models and the test dataset containing CNCPS carbohydrate fractions of ten feed mixtures with the same concentrate/roughage ratios with the dataset for modeling were used for testing the models. The performance of MLR models and BP3models were compared using the following three ways including a paired t-test, the R2and RMSPE value between the observed and predicted values. Statistical analysis indicated that VFA production (mmol) was significantly correlated with CNCPS carbohydrate fractions (g) CA, CBi, and CB2in a multiple linear pattern. Compared with MLR models, BP3models were more accurate in predicting acetate, propionate, and total VFA production while similar in predicting butyrate production. The trial indicated that both MLR and BP3models were suitable for predicting in vitro rumen VFA production of feed mixtures using CNCPS carbohydrate fractions CA, CB1, and CB2as input dietary variables while BP3models showed greater accuracy for prediction.In conclusion, this study indicated that the CNCPS carbohydrate fractions CA, CB1, and CB2were suitable dietary variables for predicting CH4, CO2, and VFA production in rumen fermentation in vitro. In addition, both MLR and BP3models were suitable for predicting in vitro rumen CH4, CO2, and VFA production of feed mixtures using CNCPS carbohydrate fractions CA, CB1, and CB2as input dietary variables while BP3models showed greater accuracy for prediction.