一維質子核磁共振代謝體圖譜之線上自動分析服務

Abstract

一維質子核磁共振圖譜(1D 1H-NMR)代謝體學的研究主要的方法之一，它擁有快速、可定量、並能提供大量檢體內化合物資訊等等特性，因此被廣泛使用。但是相同圖譜包含多種化合物的訊號，卻也造成了訊號互相重疊、分辨不易；此外它同時存在訊號變形的問題，因此核磁共振圖譜在進行統計分析之前，既定的圖譜轉換、校正等步驟是不可或缺的。為了完成這些事前處理，許多工具應運而生，遍及各種作業系統及類型，並且實作不同的功能、方法，提供此領域研究者多樣的選擇。 然而在現今的一維質子核磁共振圖譜應用工具中，大部分無法將這些處理步驟自動化，尚需使用者做部分的設定及操作，且多數不具備批次處理的功能，因此一次僅能執行一個檔案。在工具種類上，雖然至今都是以單機軟體為大宗，挾帶著免安裝、不受作業系統限制及易於擴充功能的優勢，線上工具逐漸受到重視。 我們開發並提供了一維質子核磁共振圖譜處理的線上服務，可在讀取使用者上傳的原始圖譜檔案後，經過分析處理計算檢體內的化合物成分、濃度，並輸出至檔案以供下載，這套服務除了具備自動化的特色，還支援多檔批次處理，容易上手並免除了耗費時間的人工操作。在效能表現上，我們以多個標準品的混和溶液作測試，對各個代謝物的測定顯示極高的準確度；另外也使用研究馬兜鈴對腎毒性的小鼠尿液、以及不同肌少症程度的人類尿液等複雜混和物。經由結果經過統計方法分析後，可以將樣品正確的分組，並計算出造成組間差異的主要代謝物，提供深入研究的參考。此服務無疑為這個領域開闢了新的捷徑。

Proton nuclear magnetic resonance (1H-NMR) is quantitative, rapid, reproducible technique, wildly applied in metabolomics. Capable of detecting a large number of metabolites in one measurement without sample preparation is the greatest characteristic of 1H-NMR. However, the signal overlapping and distortions are commonly seen and prevent identifications of biomarkers. To solve this issue, NMR spectrum must be pre-processed before spectra and statistical analysis in a metabolomics study. Regular NMR spectrum preprocessing procedure and statistical analysis cannot automatically process and usually requires expert’s manipulation and interpretation. Furthermore, some of them can only work on single file at once, and the others use same settings for processing multiple inputs. Since web application is platform-independent, easily-extendable, and can be used without installation, online processing tools are probably easiest for access and use. Here, we provide 1ClickNMR, a web-sever for 1D 1H-NMR data processing, which can achieve data pre-processing by auto-detecting variables or use common values, auto-deconvolute overlapping signals in NMR spectra and identify the compounds in the source mixture. It also allows multiple files in a single project and processing in one click. 1ClickNMR is easy-to-use and cost-effective with very few settings required and supports batch processing. The performance of 1ClickNMR was demonstrated with twelve mixtures composed of twelve compounds with different concentrations, mice urine samples, human urine samples and human plasma samples. The results of 12 mixtures with given concentrations were compared with the most commonly used commercial software Chenomx Suite for 1D 1H-NMR metabolomics studies. 1ClickNMR can identify the compounds automatically with much higher accuracy in terms of concentration for all twelve compounds. By using some statistical tool with the results of mice urine samples from 1ClickNMR, the data could be classified correctly correspond to the experiment groups in the reference paper. Furthermore, we compared the concentration of compounds in human urine samples from 1ClickNMR with Chenomx and LC-MS. The result from 1ClickNMR is consistent to Chenomx, although it is different from LC-MS. We also compared results of human plasma samples from Bayesil, one other web-based tool which also supports to preprocess and deconvolute with 1H-NMR raw data in a single pipeline, with 1ClickNMR. With much larger reference spectra library in 1ClickNMR, the results from 1ClickNMR gives much more accurate answers than Bayesil. Comparing with Chenomx’s heavily relying on manual manipulation and Bayesil’s long computing time with much smaller built-in database, 1ClickNMR provides an simple one-click automatically solution to analyze complex 1D-NMR metabolomics analysis.