二維凝膠電泳影像中蛋白質偵測及比對技術之研究

A study of Protein Detection and Comparison Techniques of Two-Dimensional Electrophoresis Gel Images

蔡君微(Chun-Wei Tsai)

指導教授 ： 陳同孝

國立臺中科技大學/資訊學群/資訊科技與應用研究所/碩士(2005年)

摘要

蛋白質二維凝膠電泳影像 (Two-Dimensional Electrophoresis Gel image, 2DE images) 用以進行疾病致病機轉的尋找己經很多年了。為了提升偵測比對的準確性及減少人工偵測比對所需的工作時間，目前全世界有許多相關的工具軟體被開發出來，而過去我們亦曾以鍵結碼和向量壓縮技術開發蛋白質二維凝膠電泳影像偵測和比對等相關技術。
但目前這些偵測比對技術與工具的效果都有值得加強的空間。例如：在偵測方面，目前的偵測技術在執行時，系統都需要設定多個參數值。因此，使用者需要自行決定參數，對於使用者而言，並不了解系統演算法，亦無法取得一個適用的參數值。或是使用者可使用系統所提供之預設值，但預設之參數值則無法依每一個二維凝膠電泳影像自行調整。不適合的參數值將造成不恰當的偵測結果。參數值設定的因素，通常是造成微量蛋白被正確偵測率較低，以及二維凝膠電泳影像上重疊點無法被明確定義的原因之一。
除此之外，在比對方面，目前的比對技術大都假設同一組蛋白質二維凝膠電泳影像在進行電泳實驗時，環境參數相似，則所得到之電泳影像間整體亮度以及相對應蛋白質位移程度不大。因此，所設計之比對演算法則不適用於，不同組實驗所得到的二維凝膠電泳影像，或是亮度和相對應蛋白質位移程度較大的二維凝膠電泳影像。目前比對的技術在使用上，可能會有考慮不周詳或處理速度過慢的問題。
為了提升蛋白質樣本點偵測和比對的效率和品質，以協助生物學家們快速、準確得到致病機轉。故本研究計畫開發進階之二維凝膠電泳影像中蛋白質點偵測和比對相關技術，我們稱它們為「多層次影像切割合併偵測法」和「漸進式比對法」。
在第一部份的「多層次切割合併偵測法」中，我們運用不同門檻值，針對進行偵測之二維凝膠電泳影像產生多張二質化影像(binary image)，我們的偵測方法會自動偵測進行偵測影像中每一個蛋白質樣本點最適合的參數值，並自動從多張二質化影像中取得每一個蛋白質樣本點最適當的表現方式。將同一張二維凝膠電泳影像上每一個蛋白質最適當的情況予以合併，以取得二維凝膠電泳影像中蛋白質點的最佳的表達情況，此合併之表達情況即為偵測結果，多層次影像切割合併偵測法所得到之最高精確率（precision rate）可達95%，且平均精確率 (precision rate)達89.6%，相較於IM的平均準確率76.5%明顯高出約15個百分點。
在第二部份的「漸進式比對法」研究中，我們利用多層次影像切割合併偵測法在合併過程中之二質化影像，設計階段性之比對流程。在進行比對之前，我們重新規劃蛋白質樣本點的量化特徵值及，改良傳統蛋白質群彼此相似度之定義方式(即為歐幾里德距離計算公式)設計適當之評估函式，用以完整表達蛋白質樣本點座標方面之特質。可想像成，先將影像上的蛋白質點依濃度進行分群，在相同濃度之蛋白質點群集中，依據我們的評估函式進行蛋白質點間之相似度相似度計算，以取得各濃度群集中的比對結果，最後再將各階段比對結果予以合併。經漸進式比對後兩比對影像可得到更好的比對結果，漸進式比對法之最高準確率 (precision rate)可達97%，且平均精確率 (precision rate)約94.2%，相較於IM的平均準確率89.1%亦是明顯高出許多。

關鍵字

電泳影像 ； 偵測 ； 比對 ； 微量蛋白 ； 重疊

並列摘要

The two-Dimensional electrophoresis gel image (2DE image) is important and popularly used in pathogenesis research. In our previous works, we used techniques such as chain code and vector quantization compression to develop detection and comparison techniques for 2DE images. There are also many other commercial software tools available for the same purpose. However the detection and comparison results of these tools and our past techniques cannot meet the accuracy requirement needed in protein research. For example most techniques require many parameters settings and cannot be tailor-adjusted to match the characteristics of individual protein spot. Since each individual protein spot is characteristically different, the detection process will result in some inaccuracies. Furthermore when comparing two 2DE images, the parameters used in both are quite close and are likely to be identified as images from the same group. Therefore for images with closer characteristics but are from different groups and for images with characteristics that are far apart, the current comparison techniques will result in issues such as incompleteness or slower process time.
For the biologist to be able access 2DE images quickly and accurately, it would be desirable to make improvements to increase the effectiveness and quality of the detection and comparison techniques. The proposed research is to develop improved detect and compare techniques for the 2DE images. The proposed techniques are named multi-layer cut and merge detection (MLCM) and progressive comparison techniques (PCT), respectively.
In the MLMC technique different thresholds are used on the same 2DE image to generate multiple copies of the 2DE binary images. Then on each generated image the detection technique detect for the most optimal parameters for the individual protein spots. The multiple images will then be stacked and the results from the individual generated images will be used to get optimal final detection for the overlapped protein spots. Experimental results on the MLCM technique proved that the detection technique achieved the highest detection accuracies as 95% and average detection accuracies as high as 89.6%.
In the PCT, the generated images from the MLCM technique will be used together with the concentration levels of the individual protein spots. First the proteins are quantized progressively by using a modified form of Euclidean distance equation to calculate the distance of overlapping protein spots between two images. The results are then progressively compared between every two images to get the optimal information on the protein spots. Results showed that more protein spots were can be identified. The highest comparison accuracies were as high as 97%, and average comparison accuracies as 94.2%.

並列關鍵字

Electrophoresis Gel Image ； detection ； comparison

參考文獻

[1] Andrew, S. D., Kuiper, M. J., Wride, M. A., Mansergh, F. C., Rancourt, D. E., and Mulligan, L. M. “Structural Variation in a Novel Zinc Finger Protein and Investigation of Its Role in Hireshsprung Disease,” Vol. 16, 69-76, 2003.

[2] Bledi, Y., Alex I., and Linial, M. “Proceed: A Proteomic Method for Analysing Plasma Membrane Proteins in Living Mammalian Cells,” Briefings in Functional Genomics and Proteomics, Vol. 2, No. 3, 254-265, October, 2003.

[3] Bader, G. D., Heilbut, A. H., Andrews, B., Tyers, M., Hughes, T., and Boone, C. “Functional Genomics and Proteomics: Charting a Multidimensional Map of the Yeast Cell,” Trends Cell Biol. Vol. 13, No. 7, 344-356, 2003.

[5] Chen C. H., Liu, B. D., and Yang, J. F. “Direct Recursive Structures for Computing Radix-r Two-Dimensional DCT/IDCT/DST/IDST,” IEEE Transactions on Circuits and Systems I Regular Papers, Vol. 51, 2017-2030, October, 2004.

[6] Chen, S., and Robert, M. H. “Recursive Erosion, Dilation, Opening, and Closing Transforms,” IEEE Transactions on Image Processing, Vol. 4, 335-345, March, 1995.