Premium
Data pre‐treatment and choice of resemblance metric affect how fatty acid profiles depict known dietary origins
Author(s) -
Happel Austin,
Czesny Sergiusz,
Rinchard Jacques,
Hanson S. Dale
Publication year - 2017
Publication title -
ecological research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.628
H-Index - 68
eISSN - 1440-1703
pISSN - 0912-3814
DOI - 10.1007/s11284-017-1485-9
Subject(s) - metric (unit) , trophic level , standardization , mathematics , rank (graph theory) , euclidean distance , transformation (genetics) , statistics , data transformation , multivariate statistics , multivariate analysis , biology , ecology , combinatorics , computer science , biochemistry , data mining , operations management , geometry , gene , economics , data warehouse , operating system
Fatty acids (FA) are increasingly being used in ecology to qualitatively infer diets of consumers. Analysis of FA data requires standardization to express FAs either in mg g −1 lipids, mg g −1 tissue, or as percentages of the total mass of FAs. Additionally, various transformations [square root, arcsin, log(X + 1), log‐ratio, etc.] are often used to differentially weight the contribution of less abundant FAs. The choice of standardization unit and transformation can affect interpretations of results and ultimately our understanding of trophic relationships. Data from published feeding experiments were analyzed with visualization (i.e., nMDS) and multivariate rank‐based methods (i.e. Mantel Tests) to evaluate how choice of standardization, transformation, and resemblance metric (i.e., Euclidean distance, Bray–Curtis similarities, etc.) reflects known dietary treatment groups. Our results indicate that diet interpretations were best inferred from data standardized to the total mass of FAs quantified. We found transforming data provided only weak advantages for discriminating among diet groups. Euclidean distances between FA proportions represented the known dietary differences to a high degree, are relatively easily interpretable, are applicable to a wide variety of statistical techniques, and are thus a reasonable choice of metric when analyzing FA proportions.