Format: ORAL

Daniele Silvestro1, 2, 3

1 Department of Biology, University of Fribourg 2 Department of Biological and Environmental Sciences, University of Gothenburg and Gothenburg Global Biodiversity Centre, Sweden 3 Swiss Institute of Bioinformatics, Fribourg, Switzerland

Throughout the long evolutionary history of life, species of all kingdoms have diversified through countless environmental changes and extinction events. Since most species that lived on Earth have gone extinct, it is difficult to infer how the evolution of biodiversity and ecosystems unfolded over millions of years. Here I will present a suite of new methods to infer the spatial and temporal evolutionary dynamics across major clades in the tree of life. Analyzing the fossil record through supervised and unsupervised neural networks we infer species richness dynamics in deep time, and test hypotheses on past speciation and extinction, while correcting for the gaps and biases that characterize the preservation and sampling processes. With the integration of geological and paleontological data through Bayesian deep learning models we generate reconstructions of paleovegetation and paleotemperatures through time and space providing a more accurate environmental context around macroevolutionary hypotheses. Our results show that coupling deep learning models with evolutionary models can improve our understanding of biodiversity and its evolution across time scales.