Empirical Bayes method for reducing false discovery rates of correlation matrices with block diagonal structure

BMC Bioinformatics

Software

Clare Pacini¹  Gos Micklem¹  James W. Ajioka² 

[1] CCBI, Department Applied Mathematics and Theoretical Physics, University of Cambridge, Wilberforce Road, CB3 0WA, Cambridge, UK;Cambridge Systems Biology Centre, University of Cambridge, Tennis Court Road, CB2 1QR, Cambridge, UK;Department of Pathology, University of Cambridge, Tennis Court Road, CB2 1QP, Cambridge, UK;

关键词: Empirical Bayes;    Correlation;   

DOI  :  10.1186/s12859-017-1623-y

 received in 2016-07-05, accepted in 2017-04-01,  发布年份 2017

来源: Springer

【 摘 要 】

BackgroundCorrelation matrices are important in inferring relationships and networks between regulatory or signalling elements in biological systems. With currently available technology sample sizes for experiments are typically small, meaning that these correlations can be difficult to estimate. At a genome-wide scale estimation of correlation matrices can also be computationally demanding.ResultsWe develop an empirical Bayes approach to improve covariance estimates for gene expression, where we assume the covariance matrix takes a block diagonal form. Our method shows lower false discovery rates than existing methods on simulated data. Applied to a real data set from Bacillus subtilis we demonstrate it’s ability to detecting known regulatory units and interactions between them.ConclusionsWe demonstrate that, compared to existing methods, our method is able to find significant covariances and also to control false discovery rates, even when the sample size is small (n=10). The method can be used to find potential regulatory networks, and it may also be used as a pre-processing step for methods that calculate, for example, partial correlations, so enabling the inference of the causal and hierarchical structure of the networks.

【 授权许可】

CC BY   
© The Author(s) 2017

【 图 表 】

Fig. 3

12936_2017_2051_Article_IEq78.gif

Fig. 5