Meeting Abstract

P3.102  Sunday, Jan. 6  The use of geometric morphometrics and artificial neural networks to identify teeth to species in requiem sharks (Carcharhinus sp.) SODA, KJ*; SLICE, DE; NAYLOR, GJP; Florida State Univ.; Florida State Univ.; College of Charleston,SC kjs11w@my.fsu.edu

Although many species of shark are identifiable based on tooth morphology, smooth continuous gradients in morphology from the front to the back of the jaw are common. Finding appropriate comparisons for isolated fossil teeth along this gradient, and thereby identifying species, can be difficult. A large pool of fossil shark teeth could contribute to research if a method for identifying species existed. This study introduces a method to identify upper jaw teeth from four extant species of requiem shark, Carcharhinus acronotus, C. leucas, C. limbatus, and C. plumbeus. For each species, the morphology of every upper jaw tooth in 15 specimens (178-217 teeth/species) was described using the coordinates of 13 landmarks. Using Procrustes analysis, the coordinates were standardized to remove location, orientation, and size. These coordinates were used to train a multilayer perceptron (MLP) to sort each tooth to species. MLPs are a class of artificial neural network where data is given to a set of nodes. These nodes pass their values to new nodes with each value weighted based on recipient. Each subsequent node sums its inputs, evaluates a function of the sum, and passes the result weighted by recipient. The final nodes represent species and the function evaluation is the probability that the tooth belongs to that species. The classification accuracy of the method was assessed using a 10-fold cross validation and a set of teeth from new individuals (5-15 individuals (68-215 teeth)/species). Both validation methods estimate the accuracy to be over 90% for all species. MLPs trained with Procrustes coordinates could be effective in identifying fossil teeth, as well as other hard structures that are distinct across taxa.