CHyMErA-Net

A deep learning framework for predicting efficient Cas12a guides. This software accompanies the paper:

Genetic interaction mapping and exon-resolution functional genomics with a hybrid Cas9-Cas12a platform
Gonatopoulos-Pournatzis T, Aregger M, Brown KR, Farhangmehr S, Braunschweig U, Ward HN, Ha KCH, Weiss A, Bilmann M, Durbic T, Myers Cl, Blencowe BJ, Moffat J. Nature Biotechnology (2020). DOI: 10.1038/s41587-020-0437-z

CHyMErA-Net is a Python software package that scores the efficacy of 39-nt Cas12a guide sequences (6 nt flanking upstream + 4 nt PAM + 23 nt guide + 6 nt flanking downstream). The model is a convolutional neural network trained using Keras (2.2.4) and TensorFlow (1.13.1), which takes as input the guide sequence as well as computed auxiliary features: secondary structure (minimum free energy) and melting temperatures.

Installation

The recommended installation method is via the conda package manager. CHyMErA-Net requires Python 3.5 or higher.

1. Install Miniconda3

2. Clone or download the CHyMErA-Net repository:

    git clone https://github.com/BlencoweLab/CHyMErA-Net.git
    cd CHyMErA-Net
   

3. Create a virtual environment using the provided environment.yml file, which contains the pinned dependencies used.

    conda env create -n cas12a -f environment.yml
   

4. Activate the virtual environment

    conda activate cas12a
   

5. Install the package

    python setup.py install
   

Manual installation

If the environment.yml is not used (step 2 above), then the ViennaRNA package must be installed manually:

    conda install -c bioconda viennarna

Then, install CHyMErA-Net as described in steps 4-5 above.

Data files

Input

A FASTA file with 39 nt guide sequences. Each guide sequence must consist of:

• 6 nt upstream flanking sequence
• 4 nt PAM sequence
• 23 nt guide sequence
• 6 nt downstream flanking sequence

Sequences cannot contain N characters.

Usage

chymeranet guides.fasta > scores.txt

The default output is a two column table:

1. ID of the guide sequence (taken from the FASTA file)
2. Prediction score between 0 and 1, where 0 is not effective and 1 is highly effective.

For additional options, open the help message by calling chymeranet -h.

Model files

The trained models are saved in the chymeranet/data directory:

1. The CNN Keras model
2. Scikit-learn scaling model: contains the scaling factors used in the training data