Data Organization

Switch language: 数据与目录组织

Data and codes in the original DeepChrInteract project were organized around a single entry script, preprocessing helpers, model definitions, training, testing, logged results, and sequence resources.

The current project keeps the same overall scientific workflow but restructures the repository for maintainability.

Repository structure

.
├── src/
│   ├── config.py
│   ├── dataset.py
│   ├── encoders.py
│   ├── train.py
│   ├── evaluate.py
│   └── models/
├── scripts/
│   ├── preprocess.py
│   ├── run_experiment.sh
│   └── test_pipeline.py
├── results/
├── latex/
├── DeepChrInteract-main(old)/
├── README.md
├── PRD.md
└── TASK.md

Raw data layout

The new preprocessing entry expects four text files per cell type:

data/raw/{cell_type}/
    seq.anchor1.pos.txt
    seq.anchor2.pos.txt
    seq.anchor1.neg.txt
    seq.anchor2.neg.txt

Processed data layout

After preprocessing, the repository stores split datasets as:

data/{cell_type}/
    train.npz
    val.npz
    test.npz

Each NPZ file stores:

• seqs_e: enhancer sequence strings;

• seqs_p: promoter sequence strings;

• labels: binary interaction labels.

Results layout

results/{exp_id}/seed{n}/
    config.json
    best.pt
    last.pt
    history.json
    metrics.json
    roc_curve.png
    pr_curve.png

Legacy archive

The original Keras-based implementation remains under DeepChrInteract-main(old)/. It is preserved for historical comparison, method tracing, and reference to the original writing and diagrams.