考虑多尺度特征的固有不规则蛋白质预测方法

Prediction of intrinsically disordered proteins using multi-scale analysis

固有不规则蛋白质中长、短不规则结构区域具有明显不同的氨基酸组成,现有的单一预测模型很难同时兼顾两种不规则区域的特征,导致短的不规则区域预测精度降低.为了获取更全面的特征信息,提出一种新的建模方法.针对不规则区域的序列特点,对氨基酸序列进行不同尺度的特征提取并构建基模型,采用双错测度法度量基模型间的差异度,挑选具有较大差异度的基模型进行融合,建立集成预测模型,将集成预测模型的预测结果与窗口中心三肽氨基酸形成不规则结构的统计概率相结合作为最终的预测结果.实验结果表明,所提出的方法能够有效地兼顾到长、短不规则结构区域的特征,大幅提高不规则结构区域的预测精度.

In the intrinsically disordered proteins, there are obviously different amino acids in the long and short disordered regions, while existing single prediction models （predictors） are hard to simultaneously consider the fea- tures of these two disordered regions, so that the prediction precision in short disordered region is reduced. In order to obtain more comprehensive feature information, a new modeling method was proposed to predict disordered re- gions in proteins. According to the sequence characteristics of disordered regions, the features were extracted by multi-scale analysis method and based on this the base models were built. Then the double-fault method was intro- duced to measure diversity among the base models. Some base models with bigger diversity were chosen and then fused to build the integrated predictor. Finally, the prediction outputs of the integrated predictor and the statistics probability of disordered structure formed from tripeptide amino acid in the center of window were combined as the final predicting results. The predicting results suggest that the proposed method has effectively considered the fea-tures of both long and short disordered regions and can get a good predicting accuracy.