Clustering of time-course gene expression data using a mixed-effects model with B-splines
Clustering of time-course gene expression data using a mixed-effects model with B-splines
复制标题
DOI：
10.1093/bioinformatics/btg014
复制
发表时间：
2003-03-01
期刊：
影响因子：
5.8
通讯作者：
Li, HZ
中科院分区：
文献类型：
--
作者：
Luan, YH;Li, HZ
动机：通常测量时程基因表达数据来研究动态生物系统和基因调控网络。为了解释基因表达测量随时间的时间依赖性以及微阵列数据的噪声性质，引入了使用 B 样条的混合效应模型。本文在分析时程基因表达数据和在混合模型框架中进行基因聚类时进一步探索了这种混合效应模型。结果：使用EM算法将混合模型拟合到混合效应模型框架中后，我们获得了每个簇的平滑平均基因表达曲线。对于每个基因，我们结合来自该基因和同一簇中其他基因的数据，获得了其基因表达轨迹随时间的最佳线性无偏平滑估计。模拟数据表明，这些方法可以有效地将噪声曲线聚类成曲线形状或达到曲线峰值的时间不同的簇。我们通过根据细胞周期基因表达数据对酵母基因进行聚类，以及根据成纤维细胞对血清的时间转录反应对人类基因进行聚类，进一步证明了所提出的方法。对于细胞周期调节基因的不同簇，观察到清晰的周期性模式和不同的峰值时间。人类成纤维细胞数据的分析结果显示了七种具有生物学相关性的不同转录反应谱。
Motivation: Time-course gene expression data are often measured to study dynamic biological systems and gene regulatory networks. To account for time dependency of the gene expression measurements over time and the noisy nature of the microarray data, the mixed-effects model using B-splines was introduced. This paper further explores such mixed-effects model in analyzing the time-course gene expression data and in performing clustering of genes in a mixture model framework.Results: After fitting the mixture model in the framework of the mixed-effects model using an EM algorithm, we obtained the smooth mean gene expression curve for each cluster. For each gene, we obtained the best linear unbiased smooth estimate of its gene expression trajectory over time, combining data from that gene and other genes in the same cluster. Simulated data indicate that the methods can effectively cluster noisy curves into clusters differing in either the shapes of the curves or the times to the peaks of the curves. We further demonstrate the proposed method by clustering the yeast genes based on their cell cycle gene expression data and the human genes based on the temporal transcriptional response of fibroblasts to serum. Clear periodic patterns and varying times to peaks are observed for different clusters of the cell-cycle regulated genes. Results of the analysis of the human fibroblasts data show seven distinct transcriptional response profiles with biological relevance.