以三維regulog之方法預測蛋白質與DNA之交互作用及模型

Abstract

蛋白質與 DNA 的交互作用在生物體內的機制中扮演非常關鍵的角色，這些機制包括:基因轉錄、基因重組、基因複製及DNA修復。尋找可能的蛋白質與DNA之結合配對有助於了解細胞中的調控網路，而這也是後基因體時期的重要工作。利用實驗方法來找尋可能的配對通常是昂貴且耗時的，因此我們提出一個三維regulog 的方法來預測可能的結合配對。我們的方法並提供配對的結合模型及交互作用的胺基酸及核苷酸。 我們提出一個新的計分方法來搭配三維regulog。此計分法結合了與DNA交互作用的殘基 (residue) 之演化保留以及胺基酸及核苷酸之結合傾向。我們的方法在辨識66個與DNA結合的蛋白質家族時有很高的準確率 (precision) 跟求全率(recall)。另一方面，使用我們的方法在預測蛋白質的熱點 (hotspot) 能量時亦有不錯的準確率。我們的方法並在多特定結合 (multi-specific) 蛋白質家族中亦有不錯的辨識度。 我們更進一步提出一個以知識為基(knowledge-based)的計分矩陣來增進原有計分方法的效能。使用此新的計分矩陣可讓我們預測蛋白質與DNA的結合力(binding affinity)，我們在多個不同的測試資料中都有到很好的表現，包括蛋白質與DNA的結晶結構、丙胺酸掃瞄 (Alanine-scaning) 及鋅指蛋白質 (zinc finger protein) 之實驗資料。 我們並用此方法掃瞄酵母菌HO基因的啟動子 (promoter)並找出可能的轉譯結合區(transcription factor binding sites)。

Protein-DNA interaction plays a key role in living organisms of many genetic activities such as transcription, recombination, DNA replication and repair. Finding binding pairs of proteins and DNA can help us to understand the regulatory pathway of a cell which is an important task of the post-genomic era. Experimental approaches for finding such pairs usually expensive and time-consuming. We propose computational approach called “3D-regulogs” to large scale infer protein-DNA binding partners by using the concept of regulog and the crystal structures of protein-DNA complex as templates. Such method also provides the binding model and interacting amino acids and DNA bases of predicted partners. The 3D-regulogs uses a scoring method which combines the evolutionary conservation of DNA-contact residues and the preference of interacting residues and nucleotides to evaluate protein-DNA binding partners. By applying the scoring method, we achieve high precision and recall for 66 families of DNA-binding domains, with a false positive rate less than 5% for 250 non-DNA-binding proteins. We also obtained high accuracy in predicting binding free energy of hotspot mutation sets. By testing the regulog mapping of multi-specific families, our method showed good performance to identify proteins with distinct DNA-binding specificity. For further enhancing the interaction term of the scoring function, we proposed a novel knowledge-based scoring matrix. By using such proposed scoring method, it achieved high correlation with binding affinities of several test sets, including complexes extracted from PRONIT, the Alanine-scanning set, and the base mutation set of zinc finger proteins. We also use the scoring method to scan promoter regions of yeast HO gene and obtained potential transcription factor binding sites.