iGEMDOCK: 整合自動分子嵌合、虛擬篩選與分析之圖形介面化藥物設計系統

Abstract

分子嵌合式虛擬篩選以及篩選後分析是電腦輔助藥物設計的過程中的兩項重要任務，此二方法的結合其可以有效地減低大量分子資料庫當中偽陽性藥物的出現，是先導藥物搜尋極為關鍵的步驟。本研究中，我們發展了名為iGEMDOCK的自動化的圖形介面藥物探索系統，整合了分子嵌合、虛擬篩選以及篩選後分析以及快速的藥物分子視覺化等功能。iGEMDOCK是目前現有眾多工具中第一個整合式的藥物設計系統。iGEMDOCK核心的分子嵌合及虛擬篩選的工具是基於穩固且發展完善的GEMDOCK。iGEMDOCK可直接經過其視覺化工具觀察GEMDICK以及篩選後分析所預測的分子位相。iGEMDOCK提供為篩選後分析的工具是透過預測的分子位相（蛋白質交互作用的觀點）以及配體描述子（配體構成分析）的描述方式以k-means及階層式分群的方式進行。配體構成描述子（AC）是我們新提出來的配體相似性描述方法，其觀念類似於蛋白質胺基酸序列組成分析法。我們用一組100個蛋白質-配體複合物驗證iGEMDOCK的分子嵌合準確性。用四組蛋白質標把混合隨機分子資料庫作為虛擬篩選準確性的驗證，四組資料包括了thymidine kinase活性配體、ER抑制劑（estrogen receptor antagonist）、ER促進劑（estrogen receptor agonist）以及人類二氫葉酸還原脢（DHFR）。實驗結果顯示iGEMDOCK仍保留了GEMDOCK的準確優勢，更增加了圖形化的友善操作介面。我們也驗證了新提出的篩選後分析方法AC：透過76個來自於六種不同種活性的藥物進行分群分析。結果顯示AC在這76個活性測分子當中的表現優於其他的方法。我們相信透過整合需以篩選與篩選後分析，iGEMDOCK將成為一個實用的藥物探索系統。

Structure-based virtual screening and post-screening analysis are emergent tasks in computer-based drug discovery. Combining these two methods to effectively reduce the false positives from a large compound database is considered as a key step to finding the lead compounds. In this study, we have developed a graphical-automatic drug discovery system, called iGEMDOCK, for integrating docking, screening, post-analysis, and visualization. To our best knowledge, iGEMDOCK is the first system for this requirement. The core of iGEMDOCK is the GEMDOCK, which is a robust and well-developed tool. Using iGEMDOCK, the predicted poses generated from the GEMDOCK are able to be directly visualized by a molecular visualization tool and analyzed by post-analysis tools. iGEMDOCK provides the post-analysis tools by using k-means and hierarchical clustering methods based on the docked poses (i.e. protein-ligand interactions) and compound properties (i.e. atomic compositions). Atomic composition (AC), which is similar to the amino acid composition of a protein sequence, is a new concept for measuring compound similarity. We validated the protein-ligand docking accuracy and screening accuracies of iGEMDOCK by using a test set with 100 protein-ligand complexes and four targets, respectively, which are thymindine kinase, estrogen receptor for antagonists and agonists, and human DHFR. Experimental results show that iGEMDOCK keeps the advantages of GEMDOCK and provided graphical-integrated environment for virtual screening and docking. We also evaluated the AC method on a test set with 76 compounds. The results indicate that the AC method performs better than the comparative methods in this set. We believe that iGEMDOCK, which integrates the structure-based virtual screening and post-screening analysis, is a useful system for drug discovery.