應用類神經網路分析質譜以決定胺基酸的一級結搆

Abstract

由於使用胜□質譜判斷胺基酸序列，並非簡單的工作，因此使用電腦程式 來輔助判斷，為吾人努力的方向。我們提供了一個電腦輔助判斷程式，能 夠快速、簡易的判斷胺基酸序列。本研究所使用的電腦程式為使用了類神 經網路（Artificial Neural Networks）中修正式內連結雙向關連記憶（ Modified Intraconnection Bidirectional Associative Memory）網路 的原理，提供了一個全新的電腦分析的方法。此方法是將18種胺基酸的理 論質譜碎片值經計算後視為網路的長期記憶（Long Term Memory）。我們 在判斷胜□質譜的方法，是先將樣品的連結掃描質譜（Linked Scan Mass Spectrum）轉換為程式的輸入向量，此程式可自動計算其與18種理 論胺基酸向量的相似性，進而辨識此一胺基酸殘基的種類，如此可以依序 得到可能的胺基酸序列。由於各胜□質譜訊號的相似程度相當高，因此本 程式尚無法完全正確地分析出所有的胺基酸序列，但和目前已有的判斷程 式比較，已經顯現出很大的潛力。本實驗室正致力改進實驗方法及數據前 處理過程，以期證明此一新方法的可行性及實用性。 In order to deduce the primary structure of peptide from the fast-atom-bombardment mass spectra, we developed computer programs using a modified intraconnection bidirectional associated memory (MIBAM) neural network. The calculated fragment ion masses of eighteen amino acids were employed to generate the long term memory of the network. The actual determination of sequencing was done firstly by converting the mass spectra of sample into the input vector readable to the program. The program would then produce an output vector that would be compared with the theoretical vectors of the eighteen amino acids. The best match of the comparison was suggested as the right amino acid. The same procedure was repeated to obtain the complete the sequence of peptide molecules. At the present time, we are still unable to correctly analyze the sequence for all spectra obtained in our laboratory. However, this program has already demonstrated a great potential to suppress other existing program in the literature . Future studies of the programs and mass spectral technique are under way in this laboratory to improve the performance of this novel idea.