基于结构分类的蛋白质折叠模式识别方法

PROTEIN FOLD RECOGNITION BASED ON STRUCTURAL CLASSIFICATION

对用于折叠模式识别的蛋白质结构数据库进行结构分类，构建了四个分类库：Ａｌ－α库，Ａｌ－β库，α／β库，α＋β库和一个总库，然后分别统计出不同灵敏度的匹配评估函数（平均势）。对不同的平均势，不同结构类型的蛋白进行的检验发现：来源于α／β库的平均势预测能力最强，来源于Ａｌ－α库的平均势预测能力最弱；对α／β蛋白的预测成功率最高，对Ａｌ－α蛋白的预测成功率最低。这与α／β蛋白结构最规则。

The prediction ability of mean force potentials from different protein structural classes was studied. We constructed four classified training databases: All-α database, All-β database, α/β database, α+β database and a general training database. The potential derived from α/β database has the strongest prediction ability, while the potential derived from All-α database has the weakest. α/β proteins can be predicted most successfully than any other classified proteins, while All-α proteins without comprising co-factors is predicted the worst.