AttCRISPR
OS requirement
The code was tested on Linux. Although without testing, it can also run on Windows (or other OS) in theory.
Note:
- We use tensorflow as the backend of Keras for testing.
- ViennaRNA is a C code library for prediction of RNA secondary structure, needs to be downloaded before installation.
The required software/packages are:
- python=3.6.11
- numpy=1.19.1
- scipy=1.5.2
- tensorflow=1.14.0
- keras=2.2.4
- scikit-learn=0.23.2
- biopython=1.71
- viennarna=2.4.5
- pandas=1.1.0
- hyperopt
Demo
If you want to train the standard AttCRISPR from scratch one way is to run as follow
python AttCRISPRDemo.py WTwhere WT can be replaced by ESP and SP, which means using the corresponding data set (wild-type SpCas9, enhanced SpCas9 and Cas9-High Fidelity, respectively). This work will take a lot of time, and that is why we suggest you customise your own script and run it in parallel.
Main files and directories description
- Train/TrainCNN.py and Train/TrainRNN.py, contain the code of the model's architecture in spatial and temporal domain respectively.
- Train/Ensemble.py contain the code of the ensemble model of spatial and temporal (with/without hand-crafted biological features).
- Train/Pipeline.py contain the code of the complete pipeline of training the standard AttCRISPR.
- Train/WTConst.py, Train/ESPConst.py and Train/SPConst.py, the detailed parameters of the corresponding dataset.
- Validation.py contain the code of validation.
- Util Tools to separate source data into different pkl.