inDecay : Predicting CRISPR-induced indels by frequency Decay

This library provides the inDecay package, the script for using our model to predict indels for your own input sequence, the script for training, finetuning the inDecay model.

Understand the workflow of inDecay

We provide a demonstrating notebook (demo/inDecay_demo.ipynb) containing the most necessary code to re-implement the inDecay work flow. You can follow the demo to get an idea of how the features were extracted and designed. It also records the a simplified inDecay model and the training process using pytorch_lightning Trainer.

 
To unlock the full power of inDecay, please follow the installation and training steps below.  

Installation

To run inDecay model, please install the package by

git clone https://github.com/StatBiomed/inDecay.git
cd inDecay

# create an new environment and install the dependencies
conda create -n inDecay python=3.10.4 pip

# install the python package
conda activate inDecay
pip install -r requirements.txt
pip install -e ./  

 

Data download

To get the data for re-producing the model or developing related tools, you can easily download the processed data via

# Enter the a path where you want to save the data: 
bash scripts/Data_download.sh

Set up PATH.py

After you have downloaded the data and install the SelfTarget toolkits, please runn the following script under the main directories.

bash scripts/setup_path.sh

Please change the directories mannually in PATH.py if you did not download them with default directorial setting !!

 

And we also encourage users to install indelgen toolkits from SelfTarget(https://github.com/felicityallen/SelfTarget).

conda activate inDecay
bash scripts/selftarget.sh

 

Predict with the specified model weights

To predict the editing profile for a collection of sequences, put all your sequence in a .txt file (e.g. INPUTE_SEQUENCES.txt below).

Under the main directory , run

python scripts/STfeatV2_predict.py -S <INPUTE_SEQUENCES.txt> -M <MODEL_WEIGHT.pt>

 

Train the model from scratch

To reproduce the result, you can Under the main directory , run

python scripts/STfeatv5_inDecay.py --experiment ST_June_2017_BOB_LV7A_DPI7 --read_cutoff 500 --Model_Class ST_DeepDecay --Data_transform interaction

Finetune model with Sanger sequencing data

 

For example, to finetune the model with livestock data, run

python scripts/STfeatv5_inDecay_mouse.py --data_archive species -G 0  -P pretrained/mESC_featv5_c20.ckpt -T 1