用人工神经网络方法预测蛋白质超二级结构

蛋白质超二级结构，即由α－螺旋和β－折叠等二级结构单元和连接短肽组成的超二级结构，是蛋白质结构研究中的一个重要层次。目前蛋白质超二级结构的预测工作尚属摸索阶段，还没有成熟的方法。人工神经网络预测方法是近年来在二级结构预测中发展起来的新方法。本文成功的将人工神经网络引入蛋白质超二级结构的预测工作中，结果表明蛋白质的超二级结构的发生与其局域的氨基酸的序列模式有重要联系，可以由蛋白质的一级结构序列预测该

PREDICTION OF PROTEIN SUPERSECONDARY STRUCTURE WITH ARTIFICIAL NEURO NETWORK METHOD

The artifical neuro network method is applied to pndict the proals supecondary structure(Motif), which is defined as the combination of secondary structural elements by short conned peptide. We find that the local anangelnent of amino aam fluency is of gutsignificance to the supereerendary structare. Therefore the motif structare can be pndied byits parry strUCtUre at a higher pndiction ratio (75%-80%) than the pndiction of secondarysthe. The backward propagation algorithm of aritifical neuro network is modal by using acoefficient ("momentum"), leading to an effat of faster convergence of weights matrix.