Genome-wide prediction of tissue-specific regulatory regions

The annotation and characterization of tissue-specific cis-regulatory elements in non-coding DNA represents an open challenge in computational genomics. Moreover this problem is strictly related to the prediction of regulatory variants associated with human diseases.

By leveraging recent studies that show that deep learning methods can predict active promoters and enhancers in specific tissues or cell lines, we developed deep feed forward neural networks trained with epi-genetic and deep convolutional neural networks trained with DNA sequence data, able to automatically tune their learning hyper-parameters, to successffuly predict regulatory regions active in specific cell lines (Cappelletti et al.).