Skip to content

README.md

Latest commit

 

History

History
115 lines (100 loc) · 3.25 KB

README.md

File metadata and controls

115 lines (100 loc) · 3.25 KB
Raw

NoMoCLIP

Interpretable Modeling of RNA–Protein Interactions from eCLIP‑Seq Profiles for Motif‑Free RBPs

image

1️⃣ Data availability

NoMoCLIP_dataset

2️⃣ Environment Setup

2.1 Create and activate a new virtual environment

conda create -n NoMoCLIP python=3.7.16 
conda activate NoMoCLIP

2.2 Install the package and other requirements

git clone https://github.com/yangyn533/NoMoCLIP
cd NoMoCLIP
python3 -m pip install --editable .
python3 -m pip install -r requirements.txt

3️⃣ Process data

3.1 Sequential encoding

python position_inf.py  --set_path <PATH_TO_YOUR_DATA>  --out_path <PATH_TO_YOUR_OUTPUT_DIRECTORY>

3.2 Structural encoding

This feature requires the RNAplfold tool, which is executed in a Python 2.7 environment.

python structure_inf.py  --set_path <PATH_TO_YOUR_DATA>  --out_path <PATH_TO_YOUR_OUTPUT_DIRECTORY>

3.3 Semantic encoding

python attention_graph.py \
  --kmer 1 \
  --set_path <PATH_TO_YOUR_DATA> \
  --out_path <PATH_TO_YOUR_OUTPUT_DIRECTORY> \
  --model_type <PATH_TO_YOUR_NLP_MODEL> \ 
  --maxlen 101 \
  --device cuda:1 \
  --device1 cuda:1 \
  --device2 cuda:1

3.4 Functional properties

For this feature, you need to use the corain.

python instinct_inf.py \
  --base_path <PATH_TO_YOUR_DATA> \
  --set_path <PATH_TO_YOUR_INTERMEDIATE_OUTPUT_DIRECTORY> \
  --out_path <PATH_TO_YOUR_OUTPUT_DIRECTORY> \
  --method_path <PATH_TO_YOUR_CORAIN_DIRECTORY> \ 
  --num 2

Note: The argument --num should be tested with all values in [2, 3, 5, 7, 10].

4️⃣ Training Process

python model_train.py \
  --base_path <PATH_TO_YOUR_DATA_DIRECTORY> \
  --set_path <PATH_TO_YOUR_FEATURE_DIRECTORY> \
  --out_path <PATH_TO_YOUR_OUTPUT_DIRECTORY> \
  --fold 5

5️⃣ Prediction

python model_predict.py \
  --set_path <PATH_TO_YOUR_FEATURE_DIRECTORY> \
  --out_path <PATH_TO_YOUR_OUTPUT_DIRECTORY> \
  --model_path <PATH_TO_YOUR_MODEL> \
  --gpu_id 1

🧬 Motif analysis

Motif extraction requires the installation of the MEME Suite package.

6.1 Sequential motifs

python seq_motifs.py \
  --layer <THE_LAYER_OF_MODEL_YOU_SELECTED> \
  --set_path <PATH_TO_YOUR_FEATURE_DIRECTORY> \
  --out_path <PATH_TO_YOUR_OUTPUT_DIRECTORY> \
  --model_path <PATH_TO_YOUR_MODEL> \
  --pwm_path <PATH_TO_YOUR_PWM_FILE> \
  --motif_size 7 \
  --gpu_id 1

6.2 Structural motifs

python structure_motifs.py \
  --layer <THE_LAYER_OF_MODEL_YOU_SELECTED> \
  --set_path <PATH_TO_YOUR_FEATURE_DIRECTORY> \
  --out_path <PATH_TO_YOUR_OUTPUT_DIRECTORY> \
  --model_path <PATH_TO_YOUR_MODEL> \
  --motif_size 7 \
  --gpu_id 1

📊 High attention regions

python high_attention_region.py \
  --set_path <PATH_TO_YOUR_FEATURE_DIRECTORY> \
  --out_path <PATH_TO_YOUR_OUTPUT_DIRECTORY> \
  --model_path <PATH_TO_YOUR_MODEL> \
  --gpu_id 1

Contact

Thank you and enjoy the tool! If you have any suggestions or questions, please email me at yangyn533@nenu.edu.cn.