CRISPR-VAE is a framework used for interpreting the decision of gRNA efficiency predictors, using an efficiency-aware sequence generator that allows low-level editing control. This repo contains the codes that are used to implement, train, and use CRISPR-VAE in Keras, including the synthetic data used in the paper : (TBD).

The codes can be easily operated with one-line command as shown below.

To get all requirements, env.yaml file can be used:

conda env create -f env.yaml

1. The main file is crispr_vae.py, which can be operated as the following:

python crispr_vae.py --trained_mdl --ready_synth --heatmaps --MSM --CAM

All options are boolean.

• --trained_mdl

  • Choose whether to load the weights of CRISPR-VAE ('1'), or train it from scratch ('0'). The default is '1'.

• --ready_synth

  • Choose whether to use existing synthetic data ('1'), or generate new ones ('0'). The default is '1'.

• --heatmaps

  • Choose whether to generate latent space structure heatmaps for confirmation purposes. The default is no ('0').

• --MSM

  • Choose whether to generate Mer Significance Maps (MSMs). The default is yes ('1').

• --CAM

  • Choose whether to generate Class Activation Maps (CAMs). The default is yes ('1').

python crispr_vae.py --CAM '0'

This will run the code according to the defualt options, but will exclude generating the CAMs.

All available outputs are located in /Files/outputs

If you find this repo useful, please include the following citation in your work.

(TBD) The current version of the paper can be found in (TBD).