BigBigTree: 基於Nextflow框架利用分群串接法建立巨量同源基因演化樹

Abstract

演化樹（phylogenetic tree）是根據不同生物間的型態、構造、生理、生態、遺傳和基因序列等特徵，將生物做系統化的分類，做成各物種間演化、親緣關係的樹狀圖，從中我們可以了解到序列間推斷的演化歷史。由於次世代定序技術及第三代定序技術的發展，越來越多的基因資料可以取得，面對龐大的資料量，甚至是最快的方法都具有挑戰性。一些重要的多基因家族（如嗅覺受體）已無法通過最準確的方法—最大似然（Maximum likelihood）來構建系統發育樹。\n\n在本研究中，我們提出了BigBigTree，透過分群串接法將問題分解為較小的問題並獨立解決。這個方法依賴於在直系同源基因的大型數據集中，進行分群的能力，每群直系同源基因都使用一種典型方法來構建演化樹，並在第二階段處理樹的上層(超級樹)，從每棵子樹中選擇每種物種的一種蛋白質序列，對來自同一物種的所有蛋白質序列進行多重序列比對，最後依其直系同源關係將序列串接起來，用於建構超級樹。這個方法的優點是我們減少了要分類的序列數量，且不會丟失資訊，因為最後的串接序列代表所有的序列。BigBigTree可以有效地處理特定於譜系的重複，但不能處理基因水平轉移，它更適合分析大的真核生物家族，如激酶或嗅覺受體。\n\n我們利用真實數據及模擬數據對BigBigTree進行評估，並與RAxML v8.2.12、RAxML-ng 及 IQ-TREE2 比較結果。在大多數情況下，BigBigTree的執行時間比RAxML和RAxML-ng快。在拓樸精度方面，BigBigTree在模擬數據上展現比其他方法更好的性能，並在實際數據中獲得與其他方法接近的精度。BigBigTree的原始碼及docker容器可在https://github.com/jmchanglabtw/bigbigtree和https://hub.docker.com/r/changlabtw/bigbigtree中取得。

A phylogenetic tree is a branching diagram based on the similarities of creatures in morphology, structure, physiology, genetics, ecology, and genetic sequence. It shows an inferred evolutionary history among sequences. Thanks to the next-generation sequencing technique and the third-generation sequencing technique, more and more sequences have become available. This overwhelming amount of data is challenging, even the fastest methods. Some important multi-genetic families like olfactory receptors have become impossible to build a phylogenetic tree with the most accurate methods like Maximum Likelihood (ML).\n\nHere we show how a simple Divide and Concatenate strategy, BigBigTree, can be applied to this problem by breaking it down into smaller problems that are solved independently. Our approach relies on the ability to identify within large dataset clusters of orthologous genes. Each group of orthologous genes is used to build a phylogenetic tree using a typical approach. The upper level of the tree (super-tree) is resolved in a second stage. One protein per species is chosen from each subtree. All proteins from the same species are aligned together. The alignment used for building the super-tree results from concatenating all these alignments, where within-species paralogues appear in the same columns, and orthologues appear in the same row. The advantage is that we reduce the number of sequences to classify without losing information as all sequences are represented in the final alignment. This approach can efficiently deal with lineage-specific duplications, but not with lateral transfers. It is better suited for the analysis of large eukaryotic families like the kinases or the olfactory receptors.\n\nWe evaluated BigBigTree in simulation and real data sets against RAxML v8.2.12, RAxML-ng, and IQ-TREE2. BigBigTree is faster than RAxML and RAxML-ng in most cases. Regarding topology accuracy, BigBigTree shows better performance than others in simulation data and gets compatible accuracy with others in real data. The source code and docker of the method are available at https://github.com/jmchanglabtw/bigbigtree and https://hub.docker.com/r/changlabtw/bigbigtree, where the latter allows users one-click installation.