利用成對組合方法檢測兩地區之微生物群落差異

Abstract

在微生物學中，多樣性通常是研究兩個或多個不同區域的微生物群落差異。在分析微生物群落差異時，研究人員會依據不同的生物分類等級(Taxonomic rank)作為研究目標。然而，在不同的生物分類等級所得到的統計分析結果很可能不一致。研究人員往往只關注在特定分類學水平上的微生物群落，而這導致一些重要的微生物群落訊息可能被遺漏或誤判。因此，本論文提出了一種新方法，利用成對組合（Pairwise combination）的概念以及置換多元變異數分析（Permutational multivariate analysis of variance, PERMANOVA）來深入了解不同微生物群落之間的差異。本文利用對數常態分布（log-normal distribution）模擬了兩個不同地區的微生物群落並使用混淆矩陣來展現先前研究方法對模擬資料的結果。基於此結果，我利用成對組合的方法來增加檢定結果，並使用校正演算法提高模型的預測率。其中一個結果顯示，在 delta parameter 參數為1.6的情況下，TPR（True Positive Rate, TPR）從0.643提高至0.907。然而，這同時導致 TNR（True Negative Rate, TNR）從 0.96 降低到 0.703 。校正演算法提供了一個不同的角度去分析微生物群落資料，研究人員必須考慮這些匯總統計數據之間的權衡，並根據他們的研究問題選擇最佳標準。此分析結果只適用在此次模擬資料。模型的整體準確率還需要多不同的模擬方法加以驗證。

In microbiology, diversity is an important concept that refers to the variety of different species or ecosystems present in different region. However, analyzing the difference in microbial community at various taxonomic ranks can yield inconsistent results. Researchers may only focus on a specific taxonomic level, leading to the omission or misjudgment of important microbial community information. Therefore, this paper proposes a new method that utilizes the concept of pairwise combination and permutational multivariate analysis of variance (PERMANOVA) to gain insights into the differences between microbial communities. To demonstrate the effectiveness of this method, the log-normal distribution was used to simulate microbial communities in two different regions, and a confusion matrix was used to show the results of previous research methods on the simulated data. Based on the results, a pairwise combination method was used to increase the accuracy of the tests, and a correction algorithm was employed to improve the model's prediction rate. One of the simulation results shows that when the delta parameter is set to 1.6, the TPR (True Positive Rate) increases from 0.643 to 0.907. Nevertheless, it does come with the cost of reducing TNR (True Negative Rate) from 0.96 to 0.703. The correction algorithms provide a new perspective on analyzing microbial community data, researchers must consider the tradeoff between these summary statistics and select the optimal criteria based on their research questions. The results of this analysis are only applicable to this simulation data. The overall performance of the model needs to be verified through various simulation methods.