Exploring regulatory elements in low-methylated regions for gene expression prediction.

Abstract

Recent studies on methylomes have indicated that low-methylated regions (LMRs) are related to potential active distal regulatory regions. To further investigate the potential relation between LMRs and gene expression regulation, we propose a penalized logistic regression model to predict gene expression directional change based on computationally analyzed transcription factor binding sites in LMRs that are distinctive between two cell types. We evaluated this approach using the whole genome bisulphite sequencing and RNA-seq data of two cell types: adipose-derived stem cells and iPSCs of adipose-derived stem cells. For Differentially Expressed (DE) genes with LMRs in their intergenic and/or genebody regions, our model obtained a 10-fold cross-validated AUC value of 0.88 for prediction of expression directional change. For DE genes with only LMRs in intergenic regions the corresponding AUC value is 0.84.