Design nonrepetitive and diverse activity single-guide RNA by deep learning

Abstract

AbstractMultiplex and precise control of the gene expression based on CRISPR/Cas9 is important to metabolic regulation in synthetic biology. However, employing single guide RNAs (sgRNAs) that possess repetitive DNA sequences and exhibit uniform activity could detrimentally affect the editing process, undermining both its stability and regulatory potential. In this study, we developed a deep generative model based on a decoder-only Transformer architecture (sgRNAGen) for thede novogeneration of a series of nonrepetitive and diverse sgRNAs with activity. To assess the quality of sgRNAs generated by sgRNAGen, we evaluated their activity by targeting essential genes, with the results indicating that 98% of the generated sgRNAs were active inBacillus subtilis. The generated sgRNAs were further validated for applications in single-gene editing, large fragment knockouts, and multiplex editing. Notably, the efficiency of knocking out long fragments up to 169.5 kb reached 100%, and targeting multiple sites allowed for the creation of strains with various combinations of mutations in a single editing. Furthermore, we developed a CRISPRi system utilizing the designed sgRNAs to regulate gene expression with desired strength and high precision. SgRNAGen offers a method for devising nonrepetitive and diverse activity sgRNAs, enhancing metabolic control and advancing applications within synthetic biology.TOC