lisa348/mine

MINE-数据汇交

MINE-Loop

Get the code

git clone [email protected]:MICL-biolab/MINE.git --depth=1

Prepare the environment

1. Need to prepare juicer_tools and cuda10.1 environment in advance
2. conda create -n MINE python=3.6
3. pip install torch==1.7.1+cu101 torchvision==0.8.2+cu101 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html
4. pip install -r requirements.txt

Generating data for training

Hi-C

1. .hic -> .txt(by juicer)

java -jar /path/to/juicer_tools.jar dump observed VC /path/to/hic int int BP 1000 /path/to/txt

2. .txt -> .npz

python txt2npy.py -i /folder/to/txt -o /folder/to/npz -r 1000

3. Generate training data(Hi-C)

python generate_train_data.py -i /folder/to/npz -o /folder/to/train -s 400 -f 2000

Epi

1. .bigWig -> .npz

python analysis_epi.py -i /path/to/bigWig -o /folder/to/epi -r 1000

2. Combine multiple epi data to generate correlation matrix

python epi_concat.py -i /folder/to/epis -o /folder/to/train/epi -r 1000 -s 400 -f 2000

Annotation

.bigBed -> .npz

python generate_train_annotation_data.py -i /path/to/bigBed -o /folder/to/train/annotation

or peaks -> .npz

python generate_train_annotation_data_by_peaks.py -i /path/to/bigBed -o /folder/to/train/annotation

Train

1. train

CUDA_VISIBLE_DEVICES=1,2,3,4,5,6 python -m torch.distributed.launch --nproc_per_node=6 train_model.py -i /folder/to/train -o /folder/to/checkpoint

2. validate

python validate.py --train_folder /folder/to/train --model /path/to/model --results /folder/to/result

Predict & Analyse

1. We trained a model and put it in the data folder, you can use the validate.py in the data folder to predict the data
2. The result of model prediction is analyzed by Jupyter under the folder of analyze/fig2 & analyze/fig3

MINE-Density & MINE-Viewer

1. The MINE-Density calculation is under the folder MINE_Density, and we provide two levels of calculation for TAD and compartment
2. After the first step, we also provide analysis of the density and 3D structure in the analyze/fig4 and analyze/fig5 folders