Accounting for small variations in the tracrRNA sequence improves sgRNA activity predictions for CRISPR screening

CRISPR technology is a powerful tool for studying genome function. To aid in picking sgRNAs that have maximal efficacy against a target of interest from many possible options, several groups have developed models that predict sgRNA on-target activity. Although multiple tracrRNA variants are commonly used for screening, no existing models account for this feature when nominating sgRNAs. Here we develop an on-target model, Rule Set 3, that makes optimal predictions for multiple tracrRNA variants. We validate Rule Set 3 on a new dataset of sgRNAs tiling essential and non-essential genes, demonstrating substantial improvement over prior prediction models. By analyzing the differences in sgRNA activity between tracrRNA variants, we show that Pol III transcription termination is a strong determinant of sgRNA activity. We expect these results to improve the performance of CRISPR screening and inform future research on tracrRNA engineering and sgRNA modeling. ### Competing Interest Statement JGD consults for Microsoft Research, Abata Therapeutics, Servier, Maze Therapeutics, BioNTech, Sangamo, and Pfizer. JGD consults for and has equity in Tango Therapeutics. JGD serves as a paid scientific advisor to the Laboratory for Genomics Research, funded in part by GSK. JGD receives funding support from the Functional Genomics Consortium: Abbvie, Bristol Myers Squibb, Janssen, Merck, and Vir Biotechnology. JGD interests were reviewed and are managed by the Broad Institute in accordance with its conflict of interest policies.