P5-3 Sat Jan 2 Skin lipids in Burmese pythons: comparison of data analysis approaches to multidimensional data Lincoln, JM*; Bukovich, IMG; Rucker, HR; Baedke, PE; Bartoszek, I; Parker, MR; James Madison University, Harrisonburg, VA; James Madison University, Harrisonburg, VA; James Madison University, Harrisonburg, VA; James Madison University, Harrisonburg, VA; Conservancy of Southwest Florida, Naples, FL; James Madison University, Harrisonburg, VA email@example.com
Chemical signals in vertebrates are often complex blends of molecules, many of which have independent and/or synergistic effects on receiver behavior. While isolation of these chemical mixtures is relatively straightforward, meaningful, targeted analysis of the blends can be challenging. Our lab studies the chemical composition of skin lipids that serve as communication signals in many reptile species. We treat these lipid blends as multidimensional datasets to which Bayesian statistical techniques can be applied. Here, we present an analytical framework for dealing with multidimensional chemical data from Burmese pythons (Python bivittatus). In collaboration with scientists at the Conservancy of Southwest Florida, we isolated a series of skin lipid fractions from reproductive pythons caught in mating aggregations during the breeding season across two years (n=17 males; n=19 females). These pythons are a major invasive predator of concern in Florida, and discovering the chemicals comprising their mating signals is a major target for biological control. In R, we used randomForest to predict the response variable “sex” based on gas chromatograph (GC) retention times to identify informative chemical peaks. In parallel, multiple response permutation procedure (mrpp) was used to conduct a global analysis of sex differences per fraction followed by nonmetric multidimensional scaling to visualize differences. Lipid profiles varied significantly between sexes across the different fractions, with randomForest accurately predicting sex in ~80% of the fractions. By using importance plots, we have identified key GC peaks for further testing.