Relationships among ensiling, nutritional, fermentative, microbiological traits and Clostridium tyrobutyricum contamination in corn silages addressed with partial least squares regression

The objective of this work was to reduce\r\nthe predictor dimensionality and to develop a model\r\nable to forecast Clostridium tyrobutyricum contamination\r\nin corn silages. A survey on 33 dairy farms was\r\nperformed, and samples from core, lateral, and apical\r\nparts of the feed-out face of corn silage bunkers were\r\nanalyzed for chemical, biological (digestible and indigestible\r\nNDF), fermentative (pH, ammonia nitrogen,\r\nlactic acid, VFA, and ethanol), and microbiological\r\n(yeasts and molds) traits. Corn silage samples were\r\nanalyzed for C. tyrobutyricum cell and spore counts\r\nby adoption of a molecular DNA–based method. A\r\npartial least squares (PLS) regression with a leaveone-\r\nout cross-validation method was used to reduce\r\nthe dimensionality of the original predictors (n = 30)\r\nby projecting the independent variables into latent\r\nconstructs. In a first step of the model development,\r\nthe importance of independent variables in predicting\r\nC. tyrobutyricum contamination was assessed by\r\nplotting factor loadings of both dependent and independent\r\nvariables on the first 2 components and by\r\nverifying for each predictor the variable influence on\r\nprojection values adopting the Wold’s criterion as well\r\nas the entity of standardized regression coefficients.\r\nThree ensiling characteristics (bunker type, presence of lateral wrap plastic film, and penetration resistance\r\nas a measurement of the ensiled mass density),\r\na chemical trait (DM), 9 characterizations of the fermentative\r\nprofile (pH, ammonia nitrogen, acetic acid,\r\nbutyric acid, isobutyric acid, valeric acid, isovaleric\r\nacid, ethanol, and lactic acid), and 2 microbiological\r\ntraits (yeasts and molds) were retained as important\r\nterms in the PLS model. Three reduced-variable\r\nPLS (rPLS) regressions—the first based on ensiling,\r\nchemical, fermentative, and microbiological retained\r\nimportant variables (rPLSecfm); the second based on\r\nchemical, fermentative, and microbiological retained\r\nimportant traits (rPLScfm); and the last based on only\r\nchemical and fermentative retained important variables\r\n(rPLScf)—were performed. The model that best\r\nfit the C. tyrobutyricum measurements was rPLSecfm.\r\nThe rPLScfm and rPLScf models had similar regression\r\nperformances but higher mean square errors of\r\nprediction than rPLSecfm. However, all tested models\r\nseemed adequate to rank corn silages for low, medium,\r\nand high risks of C. tyrobutyricum contamination. To\r\navoid the visit on farm by trained people required to\r\nmeasure penetration resistance, the use of the rPLScf\r\nmodel is suggested as a useful tool to assess the risk of\r\nC. tyrobutyricum in corn silage.