Uncovering tissue-specific binding features from differential deep learning

Deep learning models of MEIS differential binding in mouse branchial arches, and MEF2D differential binding in three mouse tissues (cortical neurons, retina, myotubes).

You can use the code in 'main' alone to train architectures described in the paper using your own files. Instructions are provided in a Jupyter notebook within the archive.

To replicate the results in the paper download and unpack 'main', then download and unpack the other archives into the root folder of 'main'. The zips contain TF-specific scripts, datasets, and trained models. Regions in .bed format are contained in 'meis-data' and 'mef2d-data' archives, together with shuffle permutations for replicable testing. Trained Keras models, as well as hyper-parameter sets from model selection, and scripts used to train and attribute over the datasets are contained in 'meis-models' and 'mef2d-models' archives. ChIPseq_PoissonTest file contains R code used to perform the Poisson test with ChIP-seq data.