Speaker: Chris Smith, CIRES/NOAA

Abstract: Efficient population monitoring of an increasing global list of endangered species is essential for guiding limited conservation resources. Spatial genomic data—a type of "big" data—can be used to evaluate landscape connectivity and to estimate the effective size of a population and therefore may complement other approaches for population monitoring. In this EDS seminar, I will discuss the potential, and limitations of, new deep-learning tools for inferring population density and dispersal across the landscape using geographically distributed genotype data. The proposed simulation-based-inference framework can handle data types that are available for an increasing number of species and requires relatively small sample sizes and minimal fieldwork. These developments are some of the first uses of deep learning for spatial population genetics, requiring some neural network innovations and leaving room for further improvements. Last, I will share some ambitious ideas for future projects and collaborations.

Speaker Bio: Chris is a postdoc in CIRES and has a background in computational biology. He did his PhD in the Ecology and Evolutionary Biology department (EBIO) at CU Boulder studying approximate Bayesian computation methods for demographic inference, and evolutionary genetics of sunflowers. He was also a postdoc at the University of Oregon where he worked on machine learning approaches for spatial population genetics, and he is very excited about machine learning methods development for ecology and environmental science applications.