搭配頻寬估測方法在核密度估測液相層析質譜儀資料之滯留時間校正與一維質子核磁共振代謝體圖譜之相位校正

Abstract

本篇論文呈現兩套我們研發的演算法，用來解決偵測小分子訊號時面臨的計算問題，它們是由代謝體的應用發展而來。 在本篇論文的第一個部分，我們發展一套用於液相層析質譜儀之訊號滯留時間校正工具 – LAKE，它可以校正層析質譜訊號的滯留時間(retention time)。代謝體學(metabolomics)分析上常用層析法為高效液相層析儀，因其具有系統穩定性佳、分析結果再現性高之優點，但因梯度沖提時易發生滯留時間偏移，倘若樣本內包含多元化合物，層析圖譜之滯留時間偏移將導致化合物之辨識錯誤率提高，另外現有滯留時間校正工具仍無法有效處理多批次資料之滯留時間校正，因此出現滯留時間未校正(misalignment)之情形發生，因此需要發展可處理多批次資料之滯留時間校正工具。 LAKE將偵測到之波峰資料依照樣本之資料相似性由高到低依序進行滯留時間之校正，在每一輪校正過程中會將波峰依序從質荷比(m/z)到滯留時間進行分組，再對分組完的質荷比-滯留時間群(m/z-RT group)內的滯留時間做頻寬選擇(bandwidth selector)，並使用估計出的頻寬對該組資料之滯留時間進行核密度估計(kernel bandwidth estimator)，作為滯留時間再分組的根據。每一輪結束後都會將各波峰之質荷比以及滯留時間更新為該組的平均質荷比以及平均滯留時間。LAKE可應用於在外生性化合物混和樣本，外生性化合物添加於體液樣本，以及含有多種複雜的內生性化合物樣本訊號於多批次資料之滯留時間校正。 在本篇論文的第二個部分，我們發展一個自動化的一維質子核磁共振圖譜(1D 1H-NMR)相位校正(phase correction)演算法 – PHASION，它能夠自動將多筆一維質子核磁共振圖譜完成圖譜相位校正。 在一維質子核磁共振圖譜的相位誤差來自於機器本身，是一種不可避免的誤差，需要再進行後續處理前解決此誤差消除，將頻譜還原。目前大部分研究人員所使用的圖譜相位校正方法多仰賴有經驗之使用者，依照手動方式調整參數，以求得各圖譜之最佳相位校正結果。由於這些校正方法，很容易會因為人為的因素而產生不同的校正結果，並造成後續處理結果上的差異，我們為求達到自動化的需求並且可客觀的校正一維質子核磁共振代謝體圖譜的相位，在此提出此新的自動化相位校正方法。 PHASION藉由選擇穩定訊號的圖譜區段，計算該區段圖譜在相位校正後之基線穩定程度作為分數，並搭配Nelder-Mead Simplex Optimizer最佳化搜尋方式，求出該圖譜之最佳校正之相位角度。

This dissertation presents two developed algorithms for solving computational problems of detecting small molecules in the field of metabolomics analysis. In the first part of this dissertation, we present the tool – LAKE, which is a tool for detected peak alignment to align retention time for chromatographic methods coupled to spectrophotometers such as high performance liquid chromatography for metabolomics works. The existed tools for retention time correction still can’t properly aligning retention times of detected peaks from multiple batches and some detected peaks are left misalignment. LAKE resolves peak shifts from high data similarity to low data similarity. In each turn, detected peaks would be clustered in mass-over-charged (m/z) dimension and then retention time (RT) dimension. For each m/z-RT cluster, bandwidth used in RT density estimation with kernel density estimation (KDE) is estimated with bandwidth selector. At the end of each turn of retention time shift resolution, the m/z and RT of detected peaks would be updated with average m/z and average RT of the m/z-RT group before next turn of detected peak alignment. LAKE can be applied to aligning retention time from mixed exogenic compounds samples, multiple exogenic compounds added in biofluid samples and complicate endogenous compounds contained metabolomics samples in multiple batches. In the second part of this dissertation, we present the tool – PHASION, which is a tool for automatic phase correction on multiple 1D proton nuclear magnetic resonance (1H-NMR) spectra for metabolomics works. The phase error is an unavoidable error happened when FID signal is recorded, after Fourier transformed into spectrum mixed with phase error. The phase correction is to find zeroth-order and first-order phase error to make misphased spectrum into phase-corrected spectrum before any further data processing. Current 1D 1H-NMR phase correction methods usually require manual parameter and filter tuning by experienced users to obtain desirable results from complex metabolomics spectra – thus becoming prone to correction variation and biased quantification. We present a novel alternative method, PHASION, for automatically estimating the phase angles of 1D 1H-NMR metabolomics data. PHASION finds optimal phase angles by calculating proposed objective score for relative stable segments of spectrum and calculates the score for baseline of spectrum phased with phase angles (PH0, PH1) and approach to the optimal phase angles for the spectrum with Nelder-Mead Simplex Optimizer.