New statistical tools for quantitative fatty acid signature analysis and the development of an accompanying R package

点击翻译按钮获取中文摘要

Statistical models that yield reliable estimates of predators' diets are critical to biologists studying predator/prey relationships in marine ecosystems. Quantitative fatty acid signature analysis (or QFASA) is a relatively recent diet estimation method in which the estimated proportion of each species of prey in the diet is determined by matching fatty acids in the predator and prey using statistical methods. While QFASA validation studies have demonstrated that the method offers some clear advantages over previous diet estimation methods for predators such as seals, seabirds and sea lions, they have also revealed limitations with the technique and a need for QFASA to evolve to overcome some of the issues that are now more understood. In this proposed work, we will investigate novel statistical techniques for the revamping of QFASA. Dealing with QFASA data is challenging from a statistical point of view because both the fatty acid signatures and diet estimates are compositional and often contain zeros, precluding the use of standard multivariate analysis procedures and the usual log ratio methods for managing compositional data. Furthermore the number of sampled predators is often much less than the number of variables of interest.******New methods are also needed to handle some of the practical problems of interest associated with the large amount of fatty acid data collected from free-ranging grey seals and their prey near Sable Island. Ultimately, we want to use QFASA to gain a better understanding of individual and temporal variation in resource utilization and how changes in factors such as population density and prey availability may influence variation both within and among individuals. ******Currently QFASA tools are not widely available to scientists. A significant goal of this proposal, and one that will directly benefit Canadian researchers in this field, is the creation of a QFASA R package where existing methods and those proposed here are consolidated and made easily accessible to biologists. ******The proposed work may be summarized as the development of 1) new compositional data analysis statistical methodology to improve the QFASA diet estimation procedure, 2) new statistical tools driven by real-life QFASA data, and 3) a QFASA R package which consists of a compilation of the tools developed in 1) and 2) along with previously developed QFASA methods.**