Phyloseminar
Distance-based phylogenetic placement: from traditional distances to deep learning
Phylogenetic placement, the addition of new query sequences onto an existing backbone phylogeny, has been approached extensively using approaches such as maximum likelihood. In this talk, we explore the distance-based approaches to phylogenetic placement with two goals: (1) showing how the problem can be formulated as a least-squares problem that can be solved efficiently (in linear time), and (2) exploring both traditional and exciting new directions for computing distances. We show that the distance-based framework enables placement with or without alignments, a feature that is important in assembly-free applications of placement. We also detect how instead of using traditional Markov models of sequence evolution, we can learn arbitrary neural network models from the data using machine learning. When the sequence data have not evolved on the backbone tree using the assumed model, the automatic learning method has the potential to increase accuracy. In particular, we show that the deep learning approach applied to distance-based placement, implemented in a method called DEPP, can enable the extension of a species tree (used as the backbone) using data from one or a handful of genes, instead of genome-wide data. We study the application of this single-gene species tree extension approach to combining 16S and metagenomic data.
Slides: https://drive.google.com/file/d/1fiAl...