The importance of milk mineral composition for human health is well known and accepted. Because of that, a rapid and cost effective monitoring of milk mineral composition is an important goal to achieve. This study aimed to investigate the potential of mid-infrared spectroscopy to predict calcium, potassium, sodium, magnesium and phosphorus in milk through the use of Uninformative Variables Elimination partial least squares (UVE-PLS) procedure. Individual samples (n=246) of Holstein-Friesian, Brown Swiss, Simmental and Alpine Grey cows from different stages of lactation and parities were collected from single-breed herds. Reference analysis was undertaken with inductively coupled plasma optical emission spectrometry (ICP-OES) in accordance with standardized methods. For each milk sample MIRS analysis in the range of 900 to 5,000 cm-1 was performed. Prediction models were developed using PLS regression after UVE for each considered trait and prediction accuracy was based on leave one out cross validation (246 segments). The coefficient of determination in cross validation was greatest for phosphorus (0.72), followed by potassium (0.70), calcium (0.68) and magnesium (0.67). Sodium prediction was poor, with a very low coefficient of determination (0.46). All the prediction models were unbiased (P<0.05). The ratio of performance deviation suggested that prediction models for P can be used for analytical purposes. Calcium and potassium exhibited RPD close to the threshold and thus they could be considered for analytical purposes as well. Results from the present study demonstrated that MIRS, combined with PLS regression is useful for the acquirement of milk mineral phenotypes at population level and the use of UVE combined with PLS represent a valid approach to improve the accuracy of prediction models.
Effectiveness of mid-infrared spectroscopy to predict bovine milk minerals
Visentin G;
2015
Abstract
The importance of milk mineral composition for human health is well known and accepted. Because of that, a rapid and cost effective monitoring of milk mineral composition is an important goal to achieve. This study aimed to investigate the potential of mid-infrared spectroscopy to predict calcium, potassium, sodium, magnesium and phosphorus in milk through the use of Uninformative Variables Elimination partial least squares (UVE-PLS) procedure. Individual samples (n=246) of Holstein-Friesian, Brown Swiss, Simmental and Alpine Grey cows from different stages of lactation and parities were collected from single-breed herds. Reference analysis was undertaken with inductively coupled plasma optical emission spectrometry (ICP-OES) in accordance with standardized methods. For each milk sample MIRS analysis in the range of 900 to 5,000 cm-1 was performed. Prediction models were developed using PLS regression after UVE for each considered trait and prediction accuracy was based on leave one out cross validation (246 segments). The coefficient of determination in cross validation was greatest for phosphorus (0.72), followed by potassium (0.70), calcium (0.68) and magnesium (0.67). Sodium prediction was poor, with a very low coefficient of determination (0.46). All the prediction models were unbiased (P<0.05). The ratio of performance deviation suggested that prediction models for P can be used for analytical purposes. Calcium and potassium exhibited RPD close to the threshold and thus they could be considered for analytical purposes as well. Results from the present study demonstrated that MIRS, combined with PLS regression is useful for the acquirement of milk mineral phenotypes at population level and the use of UVE combined with PLS represent a valid approach to improve the accuracy of prediction models.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
