Train DeepAnchor with three types of data

In LoopAnchor manuscript we showed that DeepAnchor model can be trained with different type of CBSs. Here we provide the reproducible code for this part.

We included 3 cell lines (GM12878, H1, K562), and 3 different CBSs (associated with ChromHMM insulator, Cohesin ChIP-seq peaks, Cohesin ChIA-PET loops). The input files have been put under ./loopanchor/data/three_type_CBSs.

run DeepAnchor

For each cell line and each type of CBSs, follow the instructions in Pipeline.ipynb to prepare the input and run DeepAnchor.

Prepare input:

python DeepAnchor_input.py work_dir

Train:

python DeepAnchor.py work_dir train

Predict:

python DeepAnchor.py work_dir predict

You only need to specify work_dir. For example, the work_dir can be ./data/three_type_CBSs/DeepAnchor_ChromHMM_GM12878.

output

The prediction score of each sample has been stored in scored_motif.tsv, which can be found in ./loopanchor/data/three_type_CBSs/{sample_name}.