Predicting CRISPR-Cas9 off-targets

In the last decade, one of the most common methods for genetic engineering is CRISPR-Cas9. This method, based on a bacterial anti-viral system, contains a Cas9 nuclease, which cleaves double-stranded DNA, and a single guide RNA (sgRNA) sequence. The sgRNA directs the Cas9 to cut in a specific genomic site. The specificity is derived from sequence matching between the sgRNA and the genomic DNA. Nonetheless, near-match of the guide may lead to the cleavage of unplanned genomic areas, termed off-target sites (OTS). OTS are hard to predict, and most datasets contain OTS with single nucleotides mismatches rather than "bulges" – one-nucleotide gups between the sgRNA and the genomic DNA. In an article published in Nucleic Acids Research, Prof. Yaron Orenstein and Ofir Yaish present a new, extended dataset of CRISPR-Cas9 OTS with bulges, which the researchers also used to train deep neural networks to predict off-target sites.