Predicting Transcription Factor Tissue Specific Binding Profiles

Transcripiton factor binding profiles are different in different cell types, yet it is necessary to understand which binding sites are occupied by a transcription factor in the target cell type to evaluate the regulatory effects. Within the framework of the ENCODE-DREAM challenge we analysed a number of quantities affecting cell type specific binding of transcription factors (TFs) including DNase accessibility, transcriptional activity of neighbouring genes, and sequence motifs of transcription factor binding sites and motifs of cofactor binding with the objective to separate the variables characterizing the cell type and the specificity of transcription factor binding. It was found that contribution of different quantities controlling the chromatin activity depends highly on the cell type and the cells of a particular cell type must be carefully selected to be used as a training dataset to obtain a predictor for TF bindng profile in the target cell type. Tests on human TFs from ENCODE yeilded a very high cell type specific TF binding site prediction accuracy for a number of transcription factors such as CTCF and ZNF143, yet for other factors the prediction accuracy was more modest. Yet, our results, at least technically, open a way to a cell-type oriented annotation of regulatory regions. Possible applications for evolutionary and plant genomics are discussed.