發展新的多序列取樣方法評估胺基酸位置保守性以預測蛋白質－受質結合位

Abstract

預測蛋白質與受質的結合位置通常是確認蛋白質功能及藥物開發的起始點，面對數量快速增加的蛋白質結構(來自結構基因體計畫)，尋找結合位置的分析工具愈來愈重要。在現存的預測方法中，我們觀察到(1)研究者以體積大小預測蛋白質－受體結合位常是不足的；(2)研究者為改進(1)的缺失而引入胺基酸位置保守性，但當此多序列比對中少數物種佔高比例時，該多序列比對通常不能反映真實的演化保守性。在這個研究中，我們發展新的多序列取樣方法評估胺基酸位置保守性來預測蛋白質與受質的結合位，此方法稱為homogenized species-based method。這個方法均量地從多個親源由遠到近的物種取出代表性序列，並計算其在序列上演化保守性之程度。接著，將此方法與蛋白質表面凹槽體積資訊相結合，運用於蛋白質與受質結合位的預測。我們以210個蛋白質－受質複合體為測試對象，當個別使用體積資訊及演化保守性程度來預測時，其預測率分別為63.8%及70.9%。另一方面，我們將體積前三名的凹槽依照演化保守性程度重新排序後，其預測率提升至71.9%。當加入選取各物種代表性序列建立的方法運用於預測蛋白質與受質的結合位，預測率上升約4%。我們將我們的方法與Consuf-HSSP做比較，結果顯示我們的方法比Consuf-HSSP的方法有更高的預測率(各為75.2%及73.1%)。同時，我們建立網頁，提供預測蛋白質－受體結合位之服務，網址為 http://gemdock.life.nctu.edu.tw/cavity_web/。

Predicting ligand-binding sites is often the starting point for protein function identification and drug discovery. Faced with a rapidly increasing number of known protein structures (e.g., structural genomics projects), it has became more important to have analytical tools that identify binding sites. Currently, we have observed that (1) It is not enough for predicting binding based only on volume of cavities; (2) In some studies, researchers combined volume information and residue conservation of cavities. When a multiple sequence alignment (MSA) has a high proportion of sequences which are from a few species, this MSA usually could not reflect natural evolutionary conservation. In this study, we developed a new profile method to predict ligand-binding site, called homogenized species-based method. This method measured the degree of evolutionary conservation based on MSA which are homogenized by selecting representative sequences from each of multiple species. Additionally, we combined volume information and evolutionary conservation to predict ligand-binding sites. We tested four methods on the dataset of 210 protein-ligand complexes. The successful rates of using only volume information and only degree of evolutionary conservation are 63.8% and 70.9%, respectively. We re-ranked the top 3 largest cavities by degree of evolutionary conservation and the performance was 71.9%. Besides, the successful rate of our homogenized species-based method was 75.2%. We compared the prediction performance of our method with Consuf-HSSP. Our method had a better successful rate (75.2%) than Consuf-HSSP (73.1%). Additionally, we built a web server for predicting binding sites at http://gemdock.life.nctu.edu.tw/cavity_web/.