从基因表达数据中有效挖掘差异共表达双聚类——DiCluster算法

摘要

The conception of bicluster is proposed by using the approach of mining on gene set and condition set parallelly. It can find genes which are co-expression under some conditions. Traditional algorithms find biclusters from only one dataset, while it is biologically meaningful to mine among a couple of datasets. This paper proposed the DiCluster algorithm. It extended nodes with the strategy of depth-prior, and added several pruning steps to mine maximal differential co-expression biclusters effectively. The result of experiment shows DiCluster is more efficient than current algorithms. And the result is more statistically and biologically significant.%双聚类是一种可以同时在基因和条件两个维度上分析基因表达数据的方法,它可以找出在部分条件下具有相似表达趋势的基因.已有的方法都是从一个数据集中挖掘双聚类.从生物意义上分析,从不同基因表达数据集中挖掘差异表达双聚类可以发现具有生物意义的转录因子等信息.因此,提出一种挖掘不同数据集上差异共表达双聚类的算法——DiCluster.该算法采用深度优先基因扩展方法,并引入了剪枝策略,有效挖掘最大差异表达双聚类.实验结果表明,DiCluster不仅比已有算法具有更高的效率,而且挖掘出的结果具有更好的统计学和生物学意义.