论文部分内容阅读
提出一种数据挖掘方法 MMHC来求解DNA序列模体。首先使用基于种子的错配聚类形成候选模体类,然后使用基于相对熵及聚类复杂度的深度优先判定(depth first determination,DFD)算法识别真正的模体类,最后使用保守区扫描法(conservation region scanning,CRS)及最大后验概率保值过滤法(MAP value-preservation filtering,MVPF)优化模体类。在两类DNA序列数据集上,将MMHC与三种经典的模体发现方法 MEME、AlignACE和SOMBRERO进行了对比试验。结果表明:对于大多数数据集,MMHC方法无论是在发现模体的可靠性及准确性方面,还是在反映背景种类的聚类结构方面,都明显优于三种经典的模体发现方法。
Proposed a data mining method MMHC to solve DNA sequence motifs. First, seed-based mismatched clustering is used to form candidate phantom classes, and then the true phantom classes are identified using a depth first determination (DFD) algorithm based on relative entropy and clustering complexity. Finally, a conservative region scanning method (conservation region scanning, CRS) and MAP value-preservation filtering (MVPF). On two types of DNA sequence datasets, MMHC was compared with three classical motif discovery methods, MEME, AlignACE, and SOMBRERO. The results show that for most data sets, the MMHC method outperforms the three classical motif finding methods both in terms of the reliability and accuracy of discovering motifs and in the clustering structure reflecting the background types.