Methods for training and interpretation of an ensemble of neural networks for multi-task functional prediction of accessibility or histone modifications from DNA sequence.
We provide a code base for training and interpreting deep learning models for predicting cell type-specific accessibility. DeepAccess is available as a python package via PyPI and bioconda as well as a binary which includes the Python interpreter (only dependency is bedtools) and is available for Ubuntu 18.04.3 LTS.
We train a DeepAccess model on ATAC-seq from 10 cell types available at this zenodo record.
Hammelman J, Gifford DK (2021) Discovering differential genome sequence activity with interpretable and efficient deep learning. PLOS Computational Biology 17(8): e1009282. https://doi.org/10.1371/journal.pcbi.1009282