Tissue Differential Expression Prediction Of Human Gene Based On Information From Proximal Promoter

Determining how transcriptional regulatory signals are encoded in vertebrate genomes is essential for understanding the origins of multicellular complexity; yet the genetic code of the cis-regulatory remains poorly understood. Comprehensive understanding promoter's role in transcription regulation is necessary for accurately reconstructing gene regulatory networks. Therefore, we performed a research on the relation between proximal promoter and the expression level of its downstream gene across 79 human tissues.Firstly, we constructed datasets of promoters with expression information of target genes. We combined gene expression date from GNF Atlas2 and promoter sequences from DBTSS to construct a positive and a negative dataset of 2000 promoter sequences respectively for each tissue; then, we analyzed the statistical characters of sequences and elements of the promoter datasets. The result shows that the frequencies of hexamer usage and the occurrence of 7 common core promoter elements distinguish between elevated and inhibited expression. Thus, a prediction system of tissue-differential expression based on information from proximal promoters - DEPS was built.The result of five-fold cross validation across 79 human tissues showed that the average sensitivity, specificity, precision and accuracy is 76.1%, 76.5%, 76.5% and 76.3% respectively. Also, tests on 28 terminally differentiated human tissues showed DEPS performed better than the results reported by similar research abroad, which is based on putative regulatory modules. In addition, in order to determine whether the predictors trained on one tissue could provide accurate predictions on other tissues, we analyzed 28 tissue-specific models across 28 tissues and the result showed the tissue specificity of the predictors and the result also provided some clue for studies on functionally related tissues.