Biclustering of gene expression data to detect co-regulated genes is a research focus in bioinformatics. The correlated pattern biclusters (CPB) algorithm overcomes the disadvantage of pattern singleness of biclusters obtained by the existing algorithms. Based on the CPB algorithm, the improved correlation biclustering algorithm (ICBA), which can globally detect high-correlation biclusters in a gene expression dataset, is presented. Firstly, ICBA randomly generated a set of genes, named Seed, which was used to initialize the candidate biclusters. Then, the genes and conditions were alternatively optimized with Pearson correlation coefficient (PCC) and mean absolute error (MAE) respectively. In the end, through the overlapping-control calculation, the redundant biclusters were filtered off. On two datasets, GDS2267 yeast gene expression dataset and human diffuse large B-cell lymphoma (DLBCL) dataset, ICBA was compared with CPB and other wildly used biclustering algorithms, such as Cheng & Church (CC), iterative signature algorithm (ISA), and order-preserving sub Matrix (OPSM). The results show that the biclusters determined by ICBA have more common transcription factors binding sites in the corresponding promoter sequences, and they also show highly enriched gene ontology (GO) functional categories. So it can be said that the genes belonging to these biclusters are highly possibly co-regulated. Therefore, the strength and superiority of ICBA are demonstrated.%用双聚类方法在基因表达数据中检测共调控基因是目前生物信息学研究的热点之一.相关模式双聚类(correlated pattern biclusters,CPB)算法克服了已有算法检测双聚类模式单一的缺陷.作者在CPB算法的基础上,提出在数据集全局范围内检测高相关双聚类结果的算法——改进的相关双聚类算法(improved correlation biclustering algorithm,ICBA),首先随机生成Seed基因集来初始化候选双聚类,然后分别用皮尔逊相关系数和平均绝对误差,对基因集和条件集交替优化,最后通过计算双聚类之间的重叠度来过滤结果.在酵母菌数据集和人类B细胞淋巴瘤表达数据集上,将ICBA与CPB等其他算法进行比较,验证了ICBA的有效性和优越性.
展开▼
