Skip to main content
ARS Home » Northeast Area » Beltsville, Maryland (BHNRC) » Beltsville Human Nutrition Research Center » Methods and Application of Food Composition Laboratory » Research » Publications at this Location » Publication #386327

Research Project: Advanced Technology for Rapid Comprehensive Analysis of the Chemical Components

Location: Methods and Application of Food Composition Laboratory

Title: Deriving Information from Complex Data Sets: Impact of Forage on Fatty Acids in Cow Milk

Author
item Harnly, James - Jim
item Picklo, Matthew
item Bukowski, Michael
item Kalscheur, Kenneth
item Magnuson, Andrew
item Fukagawa, Naomi
item Finley, John

Submitted to: Journal of Food Composition and Analysis
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/15/2021
Publication Date: N/A
Citation: N/A

Interpretive Summary: The results of scientific studies generally result in complex data sets that require machine learning to search out meaningful patterns and convert sets of seemingly meaningless numbers into information and knowledge. Using data obtained from a project aimed at determining the impact of cow forage on milk quality (the fatty composition of milk) this paper illustrates the steps necessary to obtain meaningful results. In this study, the variation between cows is very large and must be removed before more subtle trends can be seen. This paper uses common pattern recognition programs and analysis of variance to resolve the sources of variance and detect the change in fatty acid composition resulting from the differences in forage. These techniques are available to every researcher and are necessary to obtain meaningful information from complex data sets.

Technical Abstract: Machine learning and classical statistics were used to unravel the complex data set generated by the analysis of 77 fatty acids in milk from 76 cows fed 3 forages. Multivariate analysis of variance – principal components analysis (mANOVA-PCA) was used to deconvolute the data and determine the variance contributed by the main experimental factors: forage, the arbitrary cow groups, cow-to-cow variability, and milking day. All four factors contributed significantly to the data variance at the 99% confidence level as did the cross-factor variances. Cow-to-cow variance constituted 51% of the total variance of the data set and had to be removed before the significance of the other factors could be determined. The 3 forages were then found to generate milk fatty acid profiles that were significantly different. The PCA loadings showed that the major contributors to separate clustering of the forages were the less concentrated branched chain fatty acids. Fatty acids with even numbers of carbons were more concentrated, contributed high signal levels, and high noise, but were not significant in discriminating between forages. Classical paired analysis also removed the cow-to-cow variance and allowed discrimination between forages. The results showed that forages can significantly impact the fatty acid composition of milk.