基于并行多类支持向量机的蛋白质结构预测*

在远同源检测的蛋白质结构预测方法中,基于支持向量机的方法取得了优于其他方法的高准确性,但这类方法只能完成对目标蛋白质作出是否属于特定蛋白质结构的判别,而实际应用中常需要直接给出具体的结构预测结果.提出一种基于多类支持向量机的蛋白质结构预测方法,通过采用加权一对多的多类分类方法对标准支持向量机输出结果进行综合评价,获得唯...

Protein structure prediction based on parallel multi-class SVM

SVM-based methods achieved the highest accuracy compared to other methods in protein structure prediction base on remote homology detection, but these methods could only judge if the target protein had a special structure. Presented a protein structure prediction method which could predict the unique structure class by integrating the standard SVM output using weighted one-vs-rest multi-class classification method and the algorithmic performance was further improved by including parallel computing technology. The experiments show that the method based on parallel multi-class SVM obtained higher accuracy and operation speed.