新型核序列自相关函数用于非同源蛋白质结构类预测

A novel kernel sequence auto-correlation functions applying to prediction of non-homologous protein structural classes

从天然氨基酸的50个性质参数中经主成分分析得出1种新的氨基酸描述子:氨基酸特征性质得分。并在此基础上通过定义基于向量形式的自相关函数以及引入Mercer核技术将该函数运算空间进行非线性变换,最终提出了1种新的蛋白质序列表征方法:核序列自相关函数。采用该函数对632个已知晶体结构的非同源蛋白分类研究结果表明:KSACF能够恰当提取蛋白质一级序列特征以及氨基酸残基之间隐含的内在联系,从而对不同蛋白质结构类进行准确预测。

A novel amino acid descriptor termed as principal component scores of amino acid characteristic properties （SACP） was derived from 50 chemical properties of natural amino acids by using principle component analysis approach and transformed nonlinearly via mercer kernel technique to yield a vector form-based auto-correlative function. Consequently, a novel amino acid characterization protocol was presented, i.e., kernel sequence auto-correlation function （KSACF）. KSACF was then applied to perform classification study of 632 non-homologous proteins with known structures. It was indicated that KSACF is present a good performance in characterizing primary structures for proteins and potential relationship between amino acid residues, thus able to reliably predict the different types of protein structures.