OPED: an optimal prime editing guide RNA designer based on deep learning and transfer learning

Abstract

Prime editors (PEs) are promising genome editing tools, but efficiency pre-testing of prime editing guide RNA (pegRNA) design is still laborious and time-consuming due to the lack of accurate and universal approaches. Here, we design a customized attention-based model OPED and train it using transfer learning to improve the accuracy and universality of efficiency prediction and design optimal pegRNAs. We demonstrate its powerful generalization capability across diverse published test datasets. Furthermore, we extend OPED to design optimal pegRNAs and single guide RNAs (sgRNAs) to install various ClinVar human pathogenic variants, and 28 of 30 (93.33%) target sites yield desired variants with few byproducts and practical editing efficiencies of up to 29.30%, 82.84%, and 90.05% for PE2, PE3/PE3b, and ePE systems, respectively. We construct the OPEDVar database of optimal designs from over two billion candidates for all ClinVar variants and provide a user-friendly web application of OPED for any intended edit.