Purpose of DeepCRISPR

DeepCRISPR is an integrated web tool designed to unify human sgRNA on-target efficacy prediction and off-target profile prediction in one framework by deep learning. DeepCRISPR achieved the following goals: (1) For a given sgRNA, DeepCRISPR accurately predicts its possible knockout efficacy at the corresponding DNA loci. (2) For a given sgRNA, DeepCRISPR discovers its possible genome-wide off-targets and predicts their potential off-target cleavage efficacies. (3) Visualization of the genome-wide off-target profile of a given sgRNA with Circos plot.

Using DeepCRISPR for optimized sgRNA design

To use the DeepCRISPR web tool for sgRNA design, simply:

  1. Head over to DeepCRISPR home page at
  2. Click the “Prediction” page
  3. Enter in a DNA sequence or upload a FASTA file
  4. Select a target cell types in human
  5. Put your email here, and click submit.
    There are two optional inputs for DeepCRISPR:
    (1) A length of genome sequences that contains possible sgRNA locus
    (2) A list of sgRNA sequences formated in FASTA file

    (1) The predicted on-target efficacy and anti-OT summary score for detected or given sgRNA
    (2) The list of off-targets and their predicted off-target efficacies associated with corresponding

The sequence will be scanned for possible CRISPR guides across the human genome (20 nucleotides followed by a PAM sequence: NGG). Their on-target efficacy will be predicted by our deep learning model considering the sequence information and the epigenetic information in the corresponding DNA region. For each sgRNA, possible off-target matches throughout the selected genome will be scanned. Job output will be presented on the job output page with each sgRNA linked to a list of off-targets and presented with their predicted off-target efficacies. Also the whole-genome off-target profile (with up to 4 mismatches) for this sgRNA is visualized with a Circos plot. From inner circos to outer circos, different colors (Green, Yellow and Red) represents different off-target levels (Mild, Moderate and Severe). A anti-OT summary score to indicate the overall off-target profile is also calculated for each sgRNA, with the larger value of the score, the better to avoid OT. The whole output result will also be send to the email user specified. A snapshot of an example off-target profile result is presented in the following:


  1. The maximum number of sgRNA processed in one batch mode is limited to 50.
  2. The input sgRNA in the sequence or FASTA file will be checked for its uniqueness in the whole genome. If the input sgRNA is located in multiple locus in the human genome, it will be filtered by our preprocessing step.
  3. For a given sgRNA, DeepCRISPR only lists its TOP 100 off-target locus ranked by their predicted off-target scores and presented in the webpage. The complete off-target prediction results will be send to users by email.
  4. Current version of DeepCRISPR supports the genome-wide sgRNA efficacy prediction for 13 human cell lines with required epigenetic information. While it can be easily extended to other cell types if the required epigenetic information are provided. DeepCRISPR will be constantly updated to provide more cell types.

Reference of DeepCRISPR

Guohui Chuai et al, Qi Liu, DeepCRISPR : optimized CRISPR guide RNA design by deep learning, submitted, April, 2017.


Updating logs

2016-12, DeepCRISPR alpha v1.0 ready!

2017-3, DeepCRISPR alpha v2.0 launched, with inner testing for 15000+ times.

2017-4, DeepCRISPR release v1.0 launched, with focusing on human sgRNA design for NGG PAM. Current version is designed for 13 human cell types used in our model training

2017-9, DeepCRISPR release v2.0 launched, with updated network architecture and improved prediction performance!


Contact information

For any questions, please feel free to contact:

Guohui Chuai (

Qi Liu (