The Accuracy of some Models to Estimate the Coefficients of Gas Production Test in Corn Silage

Introduction:

In vitro gas production technique is one of the methods used for evaluating ruminal fermentation kinetic of feedstuffs. In this method, the volume of gas produced during the incubation is presented as a curve. The mathematical description of gas production profile is performed by fitting data set to a nonlinear model. Recently, several non-linear models have been developed to estimate gas production profile however, some of these models are not accurate enough. Therefore, the aim of this study was to investigate the accuracy of some nonlinear models for predicting ruminal fermentation kinetic of a forage feed.

Materials and methods:

 In this experiment, corn silages (samples on 0, 30 and 60 days after ensiling) were used as fermentation substrates.  Dry matter and chemical composition (organic matter, crude protein, NDF and ADF) of the samples were determined using standard methods. The rumen fluid was obtained from three fistulated rams before the morning feeding. The collected ruminal fluids were pooled and transferred into a flask to the laboratory. The rumen fluid was filtered through four layers cheesecloth, flushed continuously with CO2 and maintained at 39oC before incubation. The rumen fluid was then mixed with buffered mineral solution at the ratio of 1:2 (V/V). Gas production technique was completed in three separate runs on three different days (each run lasted 6 days). In each run, the samples were incubated in triplicate and two vials (without the substrate) were considered as the blanks. The volume of gas produced was measured at 0, 2, 4, 6, 8, 12, 16, 20, 24, 48, 72, 96,120 and 144 hours of incubation. The prediction of the gas volume at different times of incubation was compared by four nonlinear models and results were expressed in ml per 200 mg of DM incubated. The selected models (experimental treatments) included Exponential (EXP), Fitzhugh (FZH), logistic (LOG) and Gompertz (GOM). The goodness of fit of the models were evaluated using mean square error (MSE), coefficient of determination (R2), residual mean absolute deviation (RMAD) and mean percentage error (MPE). In addition, Durbin-Watson test (DW), run test and linear regression analysis (between observed and predicted values of the gas volume at different incubation times) were used to assess the accuracy of the models in fitting the data. The estimated ruminal fermentation parameters (the asymptotic gas volume and gas production rate) and goodness of fit parameters obtained from each model (MSE, R2, RMAD and MPE statistics) were analyzed using completely randomized design.

Results:

 The studied models had no difference in terms of predicting asymptotic gas volume (A) on 0, 30 and 60 days after ensiling and the value of parameter A predicted by the models were in the range of 98.37 (for GOM model on day 0) to 76.09 (for LOG model on day 60) ml per 200 mg DM. The EXP and LOG models had the highest and lowest MSE and R2 values, respectively, indicating their lower accuracy compared with GOM and FZH models. The RMAD value was lowest in GOM and FZH models (2.591 and 2.879, respectively) and was highest in EXP model (3.807). The RMAD value is used as an indicator for evaluating the goodness of fit of models. the lower values of RMAD (closer to zero), represents a better ability of the model in fitting data. Based on these results, GOM and FZH models had a higher accuracy than EXP model in fitting data. The MPE value in the EXP model (5.527) was significantly higher than the other models (p < 0.05). In other words, the predicted values (the volume of gas produced at different times of incubation) by the EXP model were lower than the observed values (it was underestimated). Based on Durbin Watson (DW) test results, the DW statistics in the EXP, FZH, GOM and LOG models were 0.392, 0.691, 0.705 and 0.675, respectively, indicating that EXP and GOM models had the lowest and highest accuracy, respectively, in predicting the rumen fermentation kinetic of corn silage. According to the run test, all the curves in EXP model had the lowest run (3 ≥) implying a poor performance of EXP model in predicting the results. The linear regression between the observed versus predicted values (regression parameters) showed a significant difference between intercept with 0; and slope with 1 in all the studied models (p < 0.05). However, based on the goodness of fit parameters obtained from the linear regression, FZH and GOM models had a better prediction of the gas production profile.

Conclusion :

The EXP model had lower accuracy in predicting the rumen fermentation kinetic of corn silage compared with the other studied models. It is recommended that other nonlinear models be used in addition to the EXP model for investigating the ruminal fermentation kinetics of corn silage.