研發一套以與受體結合為基礎的快速虛擬篩選方法

Abstract

本計畫提出一套以與受體結合為基礎的快速虛擬篩選方法來找尋資料庫中藥物設 計所要的化學小分子。文獻顯示有超過60 種結合的方法和超過30 種評分函數。很多研 究指出為了發展精確的分子結合與高效能的虛擬篩選方法，目前方法仍有很大的改良空 間。本計畫探索現有的分子結合方法並評估它們程式的平行化能力。SODOCK 聯合 AutoDock 的結合程式是第一套使用粒子群最佳化(PSO)的方法。本計畫提出一套改良 SODOCK 的方法，稱為PSODOCK﹔它是新增使用直交表設計來對PSO 初始化的高效 能取樣，是一套高度平行化的演算法。PSODOCK 只要改良參數設定便可以有效地設計 使用有繪圖處理單元的平行化版本。 以結合為基礎的快速虛擬篩選方法已成功地應用到很多的藥物標靶，通常是在做分 子結合計算之前便先利用一些已知的特徵如結合力強、物化性質適合、生體可用性高等 將沒有希望的化學小分子從資料庫中刪除。虛擬篩選法對每一個小分子候選者進行結合 及評分計算。結合法計算小分子各種可能構形（conformation）與蛋白質受體各部位結 合，而評分函數則用來指引結合計算及預測最後構形的結合強度。很多研究已顯示同時 使用多個評分函數的一致評分法比使用單一評分函數能獲得較有效的分子結合成果。本 計畫將研發一套使用PSODOCK 為基礎並使用一致評分法的快速虛擬篩選方法從已知 的化學資料庫中來過濾所要的化學小分子。所提虛擬篩選方法的效能評估是將應用到 ZINC 資料庫來挑選化學小分子並與已發表的結果做比較。

This project proposes a fast docking-based virtual screening method to filter compounds for drug discovery process. Literature reveals that more than 60 docking programs and over 30 scoring functions have been presented. Many studies indicated that significant improvements must be achieved in order to develop highly accurate molecular docking and efficient virtual screening methods. This study investigates existing docking methods and assesses their parallelization ability of implement. The docking program SODOCK cooperated with AutoDock is the first method using particle swarm optimization (PSO). The proposed docking method bases on the improvement of SODOCK, called PSODOCK, by adding an efficient sampling method based on orthogonal array for initialization of PSO, which is a highly parallel program. PSODOCK can be implemented effectively using graphics processing units (GPU) by refining the parameter settings of PSODOCK. The docking-based virtual screening method has been developed successfully applied to a number of pharmaceutical targets. Generally, an alternative method is to eliminate unpromising compounds before docking by prescreening drug-like compounds, by filtering the dataset based on appropriate property and sub-structural features, by verifying the known interactions with the target receptor, and by performing diversity analysis. Virtual screening utilizes docking methods and scoring of each compound candidate. The docking method is based on the prediction of binding modes and binding affinities of compounds by means of docking to an X-ray crystallographic structure. The probable binding mode of a ligand to differentiate correct poses from incorrect ones is based on reliable scoring functions. Scoring functions are used to direct the docking and predict the binding affinity of the final pose. It has been demonstrated that consensus scoring is generally more effective than single scoring for molecular docking. This project would develop a PSODOCK-based virtual screening method using consensus scoring to filter compounds. The application of the proposed method is evaluated by filtering compounds from ZINC database and comparing with published results.