以多源基因體計算法檢測微生物群落間與群落內之差異

Abstract

地球上蘊含大量且多樣的微生物群落，這些群落對維持生態平衡扮演著很重要的角色。為了對環境生態有更多的認識，本研究提出了一系列不同機制的計算方法來深入探討微生物群落間與群落內之關係。 在原生環境下成長的微生物，隨著時間的累積，適合生長於某一微生物群落之物種會逐漸繁殖，反之，不適合之物種則會消失滅亡。因此，微生物群落通常可由它的物種組成來表示，正確地估量微生物群落內之物種組成為本研究之重點。再者，由於不同群落間之差異與群落多樣性，準確地辨別這些差異亦很重要且具挑戰性。但是，礙於現有方法與技術，有很大比例的微生物無法於實驗室內培養。直到1998年由Jo Handelsman所提出的「多源基因體學(Metagenomics)」，藉由直接萃取環境中微生物之DNA，進行定序並分析，使無法於實驗室中培養的微生物得以窺究。因此，本研究旨在以「多源基因體學」之方式，探討微生物群落間與群落內之關係。 我們分析了三個慣用的距離方程式以辨別不同群落間之差異，發現三者於群落分群時，其結果並無顯著差異。據此，我們增加資料正規化的前處理程序，並提出整合演化距離資訊至計算方法中。研究結果顯示，無論樣本來自真實或是模擬的微生物群，運用比序進行的正規化前處理，且整合演化距離資訊，將能顯著改進群落分群。 本研究進一步提出MetaRank，利用一系列統計假設檢定和物種相對數量來減低採樣偏差的干擾，以得到一個較好的群落內之物種分佈評估。在辨識物種分佈的相關研究中，因受限於序列比對技術，通常捨棄可觀比例的片段，但這些片段可能含有相鄰基因資訊。因此，我們重新分析這些片段，檢視其對辨識物種的意義。此外，我們亦發展出一套「多源基因體」資訊處理平台(MetaABC)，整合數種分配工具，並進行資料過濾前處理及結果校正。發現 分配工具的選擇，對物種分佈結果優劣有決定性之影響。綜合上述，本研究更進一步提出一個新穎分配方法(PhymmBLxM)，可得到更好的物種分佈評估。 應用本研究提出之演算平台與方法，可增進我們對微生物群落間與群落內差異辨別議題之認識。同時，亦可評估數種關鍵因素對辨別差異效能之影響。最後，期望本研究對於探討地球環境生態之瞭解能有所貢獻。

Metagenomics enables the study of unculturable microorganisms in their original environments. The discrimination of the composition of the metagenomes from diverse microbial communities is important and challenging. Usually, each microbial community is represented by its taxonomic composition. It is essential to accurately estimate the taxonomic composition of each microbial community. Therefore, we propose a series of computational methods that use different mechanisms to discriminate the differences between and within distinct microbial communities. To discriminate the differences between distinct communities, we started with analyzing three well-known distance functions related to the strengths and limitations in the clustering of samples. The similar but distinguishable performance in clustering accuracy motivated us to incorporate suitable normalizations and phylogenetic information into the distance functions. The results indicate significant improvement in sample clustering over that derived by rank-based normalization with phylogenetic information, regardless of whether the samples are from real or synthetic microbiomes. Inspired by the rank-based normalization, we further proposed MetaRank, which employs a series of statistical hypothesis tests and the relative species abundance to reduce the noise from sampling biases and arrive at a better taxonomic estimation. We also found that existing methods discard a considerable proportion of low similarity reads when performing the taxonomic assignment (binning) process. To overcome this limitation, we retrieved the discarded reads by using conserved gene adjacency mechanism. In addition, current binning tools do not incorporate data adjustment methods while assigning reads to their respective taxa and producing abundance profiles. Hence, we developed a single platform by integrating several binning methods coupled with data filters and normalization techniques for improving the taxonomic assignment. During the development of the platform, we observed that the binning method itself is decisive while producing the species abundance profiles. We thus proposed a novel method by integrating existing binning tools to obtain a better taxonomic estimation in metagenomic analysis. In conclusion, this study explores the influence of some important factors on discriminating the differences between and within distinct microbial communities in metagenomic analysis. With the accumulation of data from sequencing technology, our study can provide a vivid understanding of more microbial communities. Thus, the analyses presented in this thesis reinforce our understanding of metagenomics in realizing the microbial communities.