Abstract: This paper implements many trials with three clustering algorithms and three metrics in seven data sets for assessment of clustering performance with these factors. It suggests a selecting exemplar algorithm for identifying the clustering parameters. Experimental results show that the metric Root Mean Square Deviation(RMSD) is better than other metrics. The algorithm SPICKER is better than others, and the AP clustering algorithm is in the next place.

Key words: protein structure prediction, candidate structure, clustering algorithm, comparability metrics, assessment

摘要： 在7个数据集上对3种不同聚类算法与3种不同相似性度量标准的多种组合进行实验，以评估这些因素对聚类性能的影响。为便于确定聚类参数，提出一种针对蛋白质结构预测的聚类中心选择算法。实验结果表明，在3种相似性度量标准中，RMSD对于聚类的效果最好，而在3种聚类算法中，SPICKER性能最优，其次是AP聚类算法。

关键词: 蛋白质结构预测, 候选结构, 聚类算法, 相似性度量, 评估

CLC Number: 

• TP311