低同源性蛋白质结构预测

The development of human genome project calls for more rapid and accurate protein structure prediction method to assign the structure and function of gene products. Threading has been proved to be successful in protein fold assignment,although difficulties remain for low homologous sequences. We have developed a method for solving the sequence rearrangement problem in threading. By reshuffling secondary elements,protein structures with the same spatial arrangement of secondary structures but different connections can be predicted. This method has been proved to be useful in fold recognition for proteins of different evolutionary origin and converge to the same fold.

基于伪氨基酸组成的G蛋白偶联受体超家族的识别

G蛋白偶联受体(GPCRs)是人体内最大的蛋白质受体家族,在制药业中起到很大作用。G蛋白偶联受体的功能和其超家族、子家族的分类密切相关,然而目前其空间结构却很难用实验方法获得。因此,如何用计算的方法预测G蛋白偶联受体的家族和超家族是生物信息学和蛋白质科学中重要的研究内容。根据Chou提出的伪氨基酸离散模型框架,使用近似熵的概念表示G蛋白序列附加特征,构造一种新的蛋白序列表示方法。采用FKNN(模糊K近邻)分类器作为预测工具,从最新的G蛋白数据抽取全部数据,经过去除同源性处理后,构成低同源性的新测试数据集。Jackknife测试结果验证了此方法的有效性。与之前的研究结果相比,取得了最高的预测精度。结果表明,此方法处理G蛋白偶联受体有很高的实用价值。