Representing interconversions among volatile fatty acids in the Molly cow model

The Molly cow model uses fixed stoichiometric coefficients for predicting volatile fatty acid (VFA) production from the fermented individual dietary nutrient fractions of forage and concentrate. We previously showed that predictions of VFA production had large errors and hypothesized that it was due to a lack of representation of carbon exchange among VFA. The objectives of the present study were to add VFA interconversion equations based on thermodynamics to the Molly cow model and evaluate the effect of these additions on model accuracy and precision of VFA predictions. Previously described thermodynamic equations were introduced to represent interconversions among VFA. The model was further modified to predict de novo acetate, propionate, and butyrate production coefficients based on forage-to-concentrate ratios rather than discrete, fixed sets of coefficients for forage-based, concentrate-based, and mixed diets. Both the original model and the modified one were reparameterized and evaluated against a common data set containing 8 studies reporting pH, VFA concentration, and VFA production rates using isotope dilution techniques and 62 studies reporting VFA concentrations and pH. Evaluations after parameter estimation revealed that predictions of VFA production rates were not improved, with root mean squared prediction errors (RMSPE) of 77, 60, and 51% for acetate, propionate, and butyrate, respectively, for the revised model versus 75, 63, and 55, respectively, for the original model. The RMSPE for predictions of VFA concentrations were reduced from 28, 46, and 40% to 22, 31, and 26% for acetate, propionate, and butyrate, respectively, simply by rederiving the VFA coefficients, but minimal further improvement was achieved with the addition of thermodynamically driven interconversion equations (RMSPE of 21, 32, and 27% for acetate, propionate, and butyrate, respectively). Thus, the results indicate that thermodynamically driven interchanges among VFA, as represented in this study, may not be a primary determinant for the accuracy of predictions of net production rates. Including the effect of pH on VFA absorption reduced the mean bias of propionate production and slope bias of acetate production, but not the overall RMSPE. The larger prediction errors for VFA production as compared with concentrations suggest the data quality may not be high, or that our representation of VFA production and absorption as well as ruminal digestion is inadequate. Additional data are required to discriminate among these hypotheses.