DEPP: Deep learning enables extending species trees using single genes

Placing new sequences onto reference phylogenies is increasingly used for analyzing environmental samples, especially microbiomes. However, existing placement methods have a fundamental limitation: they assume that query sequences have evolved using specific models directly on the reference phylogeny. Thus, they can place single-gene data (e.g., 16S rRNA amplicons) onto their own gene tree. This practice is a proxy for a more ambitious goal: extending a (genome-wide) species tree given data from individual genes. No algorithm currently addresses this challenging problem. Here, we introduce Deep-learning Enabled Phylogenetic Placement (DEPP), an algorithm that learns to extend species trees using single genes without pre-specified models. We show that DEPP updates the multi-locus microbial tree-of-life with single genes with high accuracy. We further demonstrate that DEPP can achieve the long-standing goal of combining 16S and metagenomic data onto a single tree, enabling community structure analyses that were previously impossible and producing robust patterns.

将新序列置于参考系统发育树上的系统发育放置方法，如今已愈发广泛地应用于环境样本分析，尤其是微生物组研究。然而，现有的序列放置方法存在根本性局限：它们假定查询序列是直接基于参考系统发育树，按照特定进化模型演化而来。因此，此类方法仅能将单基因数据（如16S rRNA扩增子）放置至对应的基因树上。此类操作实则是为了实现一项更具挑战性的目标：即利用单基因数据拓展全基因组物种树，但目前尚无算法能够解决这一难题。为此，我们提出深度学习辅助系统发育放置算法（Deep-learning Enabled Phylogenetic Placement，DEPP），该算法可无需预设进化模型，仅通过单基因数据完成物种树的拓展。实验结果表明，DEPP能够以极高的准确度利用单基因更新多位点微生物生命之树。进一步研究显示，DEPP能够实现将16S rRNA扩增子与宏基因组数据整合至同一系统发育树这一长期以来的目标，从而开展此前无法实现的群落结构分析，并得到稳健可靠的分析结果。