PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Identifying targets of transcriptionally regulated transcription factors using dynamical models
Antti Honkela, Neil Lawrence and Magnus Rattray
In: Mathematical and Statistical Aspects of Molecular Biology: 19th Annual MASAMB Workshop, 2-3 Apr 2009, London, UK.

Abstract

Identifying potential targets of transcription factors (TFs) is an important first step in inferring gene regulatory networks. We apply the cascaded differential equation regulation model proposed by Gao et al. (2008) to identify targets of transcriptionally regulated TFs from expression data. The model includes ordinary differential models of both translation of the TF protein and transcription of the target genes. Setting a Gaussian process prior on the input TF expression profile leads to a joint Gaussian process over all observables. Alternative models can be scored by their marginal likelihoods. We apply the model for each candidate target individually to obtain a first ranking that can later be refined by modelling a number of likely targets jointly. Targets identified for TFs involved in Drosophila mesoderm development are significantly enriched in identified ChIP-chip binding sites, genes differentially expressed in knock-outs and genes with annotated expression in the mesoderm. We attain over 0.1 improvements in the area under ROC curve over the method of Della Gatta et al. (2008) and 0.09 improvements over ranking by differential expression in knock-outs. References P. Gao, A. Honkela, M. Rattray, N.D. Lawrence. Bioinformatics 24(16):i70-i75 (2008). G. Della Gatta et al. Genome Research 18(6):939-948 (2008).

EPrint Type:Conference or Workshop Item (Talk)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Computational, Information-Theoretic Learning with Statistics
ID Code:5952
Deposited By:Antti Honkela
Deposited On:08 March 2010