OJGG Publishes Landmark Methods Paper on Off-Target Prediction for CRISPR Prime Editing
The Onyx Journal of Genetics & Genomics (OJGG) has published a landmark methods paper describing PrimeGuard, a deep-learning framework for predicting off-target activity of CRISPR prime editors. The work, led by Dr. Mei-Lin Cho (Seoul National University) and colleagues at the Broad Institute, demonstrates a 41% reduction in false-positive off-target predictions relative to the previous state of the art across 19 human cell lines.
PrimeGuard integrates a graph-attention network over the expanded protospacer context with a thermodynamic prior on pegRNA secondary structure. The authors validated predictions experimentally using ultra-deep amplicon sequencing across more than 12,000 candidate off-target sites, achieving an area under the precision–recall curve of 0.94 on held-out guide RNAs.
All code, pretrained weights, and the full validation dataset are released under a permissive Apache 2.0 licence. To facilitate adoption by clinical gene-therapy developers, the authors have also contributed a hosted inference endpoint and a structured reporting template compatible with regulatory submissions to the EMA and FDA. The paper is available immediately at genetics.onyx-press.org.