DEEP LEARNING APPROACHES TO THE PHYLOGENETIC PLACEMENT OF EXTINCT POLLEN MORPHOTYPES

The phylogenetic interpretation of pollen morphology is limited by our inability to recognize the evolutionary history embedded in pollen features. Deep learning, however, offers tools for connecting morphology to phylogeny. We developed an explicitly phylogenetic toolkit using neural networks for characterizing the overall shape, internal structure, and texture of a pollen grain. We demonstrate that machine-learned features can be used to quantify morphological differences among pollen taxa, discover novel morphotypes, and place fossil specimens in a phylogeny. Our analysis pipeline first determines whether testing specimens are from unknown species based on uncertainty estimates. We then transform machine-learned classification features into phylogenetically-informed ones using a trained multi-layer perceptron network. These transformed features are functionally continuous characters that can be used to place specimens in a phylogeny through Bayesian inference. We trained and evaluated our models using optical superresolution micrographs of 30 Podocarpus species and then used these trained models to place nine fossil Podocarpidites specimens within a reference phylogeny. In doing so, we provide evidence that the phylogenetic history encoded in pollen morphology can be recognized by neural networks and that machine-learned features can be employed in phylogenetic placement. Our approach fundamentally changes how fossil pollen morphology can be interpreted, making extinction and speciation events that would otherwise be masked by the limited taxonomic resolution of the fossil pollen record visible to palynological analysis, thereby allowing for greater evolutionary inference from fossil pollen specimens. This approach is not limited to fossil pollen and can be generalized to other taxa and biological images.