Improved off-target prediction by incorporating chromatin data by the CROP-IT algorithm
Off-target cleavage is a big issue in genome editing, particularly with CRISPR. It seems that, although way simpler to use, CRISPR is not as specific as ZNFs or TALENs. A further problem is that predicting off-target sites is very challenging, for example in the recent report of
Tsai S et al., an unbiased screening for off-target double strand breaks revealed sites that were not predicted by the
MIT CRISPR tool or the
E-CRISPR algorithm. The prediction programs however returned several potential sites that did not appear to be off-target sites. More accurate prediction of off-target sites is crucial since most researchers base their guide RNA design of these algorithms. Since for each gene, a huge number of guide RNAs could be potentially designed, accurate prediction of off-target sites would help to score gRNAs based on specificity. At this point we are very far away from this. One major reason for low efficiency prediction is differential accessibility of the DNA depending on chromatin state. The
Adli lab (University of Virginia) now developed an improved algorithm for off-target search, the
CROP-IT, which incorporates chromatin information.
The CROP-IT allows to search for sites up to 6 mismatches (cleavage) and up to 9 mismatches (binding). This is already an important discrimination, since certain applications (e.g. epigenome editing) requires only Cas9 binding, but not cleavage. Binding or cleavage tolerates different mismatches and now this feature is built into this prediction software. The prediction scores mismatches in relation to the PAM site and then incorporates DN-ase I hypersensitivity information into this score. The PAM site relation scoring is based on training of the algorithm on available ChiP-Seq and GUIDE-Seq data. DN-ase I sensitive sites reflect more open chromatin and more accessible DNA, however this is specific for each and every type of cell. The CROP-IT uses an average from 125 different human cell lines at this point. CROP-IT seems to outperform current prediction algorithms in Cas9 binding and cleavage prediction when validated by ChiP-Seq (Cas9 binding) and cleavage (GUIDE-Seq).
It is available for S. pyogenes, NGG and NNG PAM sites are analyzed separately.
Link to the paper: http://nar.oxfordjournals.org/content/early/2015/06/01/nar.gkv575.full
Link to the CROP-IT tool: http://cheetah.bioch.virginia.edu/AdliLab/CROP-IT/homepage.html