預測蛋白質間交互作用之新能量模式

Abstract

我們發展一個新的知識為基的鍵結能量模式(Knowledge-based energy model) 和一個快速計算鍵結能量的方法，此模式可以應用在蛋白質交互作用的預測。此模式包含167 原子型態和19原子距離類別，不僅可以考慮原子和原子之間的交互作用更可以考慮胺基酸和胺基酸之間的交互作用，本論文以641蛋白質和蛋白質之間交界的結晶構形的資料庫推導蛋白質之間的鍵結能量。統計上的意義可以顯示出靜電力、氫鍵、凡得瓦力、苯環對苯環的交互作用的在物理化學上的意義。我們測試我們新的能量模式在20鍵結和20未鍵結系統上且比其他相關的能量模式更能區分正確和不正確的構形在我們的測試系統上。我們發現考慮20個胺基酸所有的原子並且把胺基酸視為唯一和19個原子距離類別的設計可以增加對鍵結親和性的敏感度。更進一步，我們證實原子之間距離類別的區間大小設計會影響到鍵結和未鍵結系統上測試的結果。測試的結果顯示我們能量模式是健全的且可以實際應用在蛋白質交互作用的預測。

We have developed a new binding energy model and a fast method calculating binding energy for predicting protein-protein interactions. This model yielded the advantages of both atom-atom and residue-residue contacts by designing 167 atom types and 19 distance classes. This all-atom residue-specific binding energy in intermolecular interactions was derived from a nonredundant dataset with 641 co-crystallized protein–protein interfaces. These statistic preferences show that this model is able to reflect physical meaning of hydrogen bonding, disulfide bonding, van der Waals contacts, electrostatic force effect, and aromatic-aromatic interactions. We tested our new model on 20 bound and 20 unbound systems and showed that the results were better than those using other models. Our energy model almost completely discriminated the native states and near-native states for 40 test systems. We found that the all-atom residue-specific with multiple distance classes increased sensitive to binding affinity and the interval size of a distance class influenced the performance for bound and unbound systems. These results suggest that our model is robust and useful for protein-protein interactions.