Data Organization

Data and codes in https://github.com/lichen-lab/DeepChrInteract are listed as follows in order to run DeepChrInteract

• DeepChrInteract

  The root folder of this project. All files should be stored in this directory and its subdirectories.

  • DeepChrInteract.py

    Main file to call all functions. Use python3 DeepChrInteract.py -h to browse all functions and options

  • data_preprocessing.py

    Used to preprocess data, generate npz and png files, use python3 DeepChrInteract.py -p true -n [file name]

  • model.py

    Stored all models, including: onehot_cnn_one_branch / onehot_cnn_two_branch / onehot_embedding_dense / onehot_embedding_cnn_one_branch / onehot_embedding_cnn_two_branch / onehot_dense / onehot_resnet18 / embedding_cnn_one_branch / embedding_cnn_two_branch

  • train.py

    Called by model.py to train the model. Use python3 DeepChrInteract.py -m [model name] -t train -n [file name]

  • test.py

    Called by model.py to test the model. Use python3 DeepChrInteract.py -m [model name] -t test -n [file name]

  • log.txt

    Used to store prediction results by test.py, Stores the source gene of the timestamp model, the test target gene, the aucroc result, and the Pearson correlation coefficient

  • embedding_matrix.npy

    Pretrained DNA2VEC embedding matrix from hg19 human genome, which contains 4097*100 dimensions (6mer, 2**6=4096, where the first line is the initial line, all 0)

  • resnet.py

    Include resnet18, resnet34, resnet50, resnet101, resnet152. This file is a resnet library file.

  • File path data

    Store DNA sequences from labelled chromatin interactions and non-chromatin interactions

    • File path Example: AD2.po

      • seq.anchor1.pos.txt

        DNA sequence for chromatin-interacted region1. Each row is a sequence.

      • seq.anchor2.pos.txt

        DNA sequence for chromatin-interacted region2. Each row is a sequence.

      • seq.anchor1.neg2.txt

        DNA sequence for non-chromatin-interacted region1. Each row is a sequence.

      • seq.anchor2.neg2.txt

        DNA sequence for non-chromatin-interacted region2. Each row is a sequence.

  • File path h5_weights

    Saved weights for neural network

  • File path result

    Consists of folders for multiple datasets and multiple model folders for each dataset