基于二级结构端点位置残基分布特异性改进二级结构预测工具的精度

Protein secondary structure prediction is one of most important problems of bioinformatics. However, almost all the secondary structure prediction methods used the same features for all the sites in protein, and they ignored the residues distribution diversity on different positions of secondary structures. Based on the residue position-specific performance, we built the prediction modes by Support vector machine (SVM) to predict the probability of one position to be secondary structure ends (i.e. helix/strand start/finish positions), and developed an end-position modification scheme to correct the ends prediction results of exist secondary structure prediction tools. Here we tested the modification scheme on two tools, psiSVM and PSIPRED, and the results showed that, compared with psiSVM and PSIPRED, the ends prediction accuracy by modification improved about 10% and 5% respectively, and the whole prediction accuracy (Q3 value) also got some improvement.