KrisprKreme

An efficient command-line tool for CRISPR-CAS9 experiment design.

Disclaimer: The indices from the human genome take approx 2 days to build, if you want the prebuilt indices contact us

ssures11@jhu.edu pvsriv@cs.jhu.edu

Usage

The following steps were successfully executed in an Ubuntu Box - 64-bit 16.04 Xenial.

Step 1: Install the tool dependencies using

$ pip install -r requirements.txt

Step 2: Download the Gemtools aligner binaries from http://gemtools.github.io and add the Gemtools' bin folder to $PATH

Step 3: Download and install the ViennaRNA package from https://www.tbi.univie.ac.at/RNA

Step 4: For the scoring Machine Learning model we have the unfeaturied data in src/model/processed.json. This has to be converted to one-hot encoded featurized csv. In order to do that run

$ python src/model/featurize.py

Note: This will take some time

Step 5: Now that the data is featurized we need to create the Random Forest model by doing

$ python src/model/scoring_model.py create_model

Step 6: To easily test out a smaller genome, we have added the phi-X virus genome to source code in Phi-Genome folder. To build Bloom and GEM indices for this

$ python krispr_kreme.py build -p Phi-Genome

Step 7: To querying step happens in 2 steps: 1. The 20nt sgRNA sequence is queried in the Bloom Filter indices of all chunks (for phi-X virus it is only one chunk but for human genome each chunk is a chromosome). 2. The sequence is then aligned using GEM indices of only the chunks obtained from the previous step because only in those chunks the sequence is found.

To do this, run
```
$ python src/krispr_kreme.py query -s "GATGCTGTTCAACCACTAAT" -p Phi-Genome -o alignments-phi.json
```

Step 8: To do on-target and off-target scoring of the output of the query and alignment steps run

$ python src/krispr_kreme.py score -s "GATGCTGTTCAACCACTAAT" -p alignments-phi.json -o output-phi.json

Machine Learning Model Evaluation

To plot the ROC and Precision-Recall curves for the SVM model and the Random Forest model run

$ python src/model/scoring_model.py plot_curves

Deep Learning Model Evaluation

To run the training/validation process of the Deep Convolution Neural Network follow these steps:

Install Caffe by following the instructions https://github.com/BVLC/caffe/wiki/Ubuntu-16.04-or-15.10-Installation-Guide. Use the CPU mode of Caffe and make sure you do export $PATH=<caffe-root-path>/build/tools/:$PATH. To train the network

$ cd src/model/cnn
$ caffe train --solver=sgrna-dcnn-solver.prototxt

Acknowledgements

1. For our relative Cutting Frequence Determination (CFD) we have used the code provided by Doench et al. Optimized sgRNA design to maximize activity and minimize off-target effects of CRISPR-Cas9 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4744125/)
2. To generate hamming neighbors we have used Prof.Benjamin Langmead's code from his homework solutions
3. For indexing and querying the genome using Bloom Filter we have used https://pypi.python.org/pypi/pybloom