Mukhopadhyay Anirban, Maulik Ujjwal, Bandyopadhyay Sanghamitra
Department of Computer Science and Engineering, University of Kalyani, Kalyani-741235, West Bengal, India.
J Bioinform Comput Biol. 2009 Oct;7(5):853-68. doi: 10.1142/s0219720009004370.
Biclustering methods are used to identify a subset of genes that are co-regulated in a subset of experimental conditions in microarray gene expression data. Many biclustering algorithms rely on optimizing mean squared residue to discover biclusters from a gene expression dataset. Recently it has been proved that mean squared residue is only good in capturing constant and shifting biclusters. However, scaling biclusters cannot be detected using this metric. In this article, a new coherence measure called scaling mean squared residue (SMSR) is proposed. Theoretically it has been proved that the proposed new measure is able to detect the scaling patterns effectively and it is invariant to local or global scaling of the input dataset. The effectiveness of the proposed coherence measure in detecting scaling patterns has been demonstrated experimentally on artificial and real-life benchmark gene expression datasets. Moreover, biological significance tests have been conducted to show that the biclusters identified using the proposed measure are composed of functionally enriched sets of genes.
双聚类方法用于在微阵列基因表达数据的实验条件子集中识别共同调控的基因子集。许多双聚类算法依靠优化均方残差来从基因表达数据集中发现双聚类。最近已证明，均方残差仅擅长捕获恒定和移位双聚类。然而，使用此度量无法检测到缩放双聚类。在本文中，提出了一种称为缩放均方残差（SMSR）的新一致性度量。理论上已证明，所提出的新度量能够有效检测缩放模式，并且对输入数据集的局部或全局缩放具有不变性。所提出的一致性度量在检测缩放模式方面的有效性已在人工和实际基准基因表达数据集上通过实验得到证明。此外，还进行了生物学意义测试，以表明使用所提出的度量识别出的双聚类由功能丰富的基因集组成。
