DEPP: Deep Learning Enables Extending Species Trees using Single Genes

Abstract

Identifying samples in an evolutionary context is a fundamental step in the study of microbiome, and more broadly, biodiversity. Extending a reference phylogeny by placing new query sequences onto it has been increasingly used for sample identification and other applications. Existing phylogenetic placement methods have assumed that the query sequence is homologous to the data used to infer the reference phylogeny. Thus, they are designed to place data from a single gene onto a gene tree (e.g., they can place 16S sequences onto a 16S gene tree). While this assumption is reasonable, ultimately, sample identification is a question of identifying the species not individual genes. The placement of single gene data on a gene tree is therefore used as a proxy for a more ambitious goal: extending a species tree given sequence data from one or more gene. This goal poses difficult algorithmic questions. Nevertheless, a sufficiently accurate solution would not only improve sample identification using marker genes, it would also help achieving the long-standing goal of combining 16S and metagenomic data. We approach this problem using deep neural networks (DNN) and introduce a method called DEPP. Given a reference species tree and sequence data from one (or a handful of) genes, DEPP learns how to extend the species tree to include new species. DEPP does not rely on pre-specified models of sequence evolution or gene tree discordance; instead, it uses highly parameterized DNNs to learn both aspects from the data. We test DEPP both in simulations and on real microbial data and show high accuracy.

Fast and Accurate Distance-based Phylogenetic Placement using Divide and Conquer

“…In this framework, the input to APPLES is a reference (a.k.a backbone) phylogenetic tree T with n leaves and a vector of distances δ qi between a query taxon q and every taxon i on T . Although machine-learning based methods shows substantial promise (Jiang et al, 2021), typically, input distances are obtained by calculating sequence distances between query and backbone taxa followed by a phylogenetic correction using a statistically consistent method under a model such as Jukes-Cantor (JC69) (Jukes and Cantor, 1969). APPLES introduced a dynamic programming algorithm to find a placement of q that minimizes weighted least squares error n i=1 w qi (δ qi − d qi (T )) 2 where d qi (T ) represents the path distance from q to backbone taxon i on T .…”

“…A main attraction of placing new sequences onto an existing phylogeny is computational expediency: the running time of phylogenetic placement is a fraction of the time needed for de novo reconstruction and can grow linearly with the number query samples assuming they are placed independently. To take advantage of this potential, many methods have been developed using a wide range of algorithmic techniques (e.g., Barbera et al, 2019;Brown and Truszkowski, 2013;Jiang et al, 2021;Linard et al, 2019;Matsen et al, 2010;Mirarab et al, 2011;Rabiee and Mirarab, 2018;Stark et al, 2010;Zheng et al, 2018).…”

<strong></strong> Recent Progress on Methods for Estimating and Updating Large Phylogenies

“…Given an existing species tree T , INSTRAL will add the new species into the existing tree to optimize the quartet tree support for the extended species tree (i.e., INSTRAL extends the theoretical approach in ASTRAL). Another new method is DEPP (Jiang et al, 2021), which computes distances using a deep neural network (DNN) and then runs APPLES to place the new species into the tree. By training the DNN appropriately, these distances can be appropriate to this problem of adding species into species trees.…”

Fast and accurate distance‐based phylogenetic placement using divide and conquer

“…To take advantage of this potential, many methods have been developed using a wide range of algorithmic techniques (e.g. Balaban & Mirarab, 2020; Barbera et al, 2019; Brown & Truszkowski, 2013; Jiang et al, 2021; Linard et al, 2019; Matsen et al, 2010; Mirarab et al, 2011; Rabiee & Mirarab, 2018; Stark et al, 2010; Zheng et al, 2018).…”

“…a backbone) phylogenetic tree T with n leaves and a vector of distances between a query taxon q and every taxon i on T . Although machine learning‐based methods show substantial promise (Jiang et al, 2021), typically, input distances are obtained by calculating sequence distances between query and backbone taxa followed by a phylogenetic correction using a statistically consistent method under a model such as Jukes–Cantor (JC69) (Jukes & Cantor, 1969). APPLES introduced a dynamic programming algorithm to find a placement of q that minimizes weighted least squares error where represents the path distance from q to backbone taxon i on T .…”