Design of custom CRISPR-Cas9 PAM variant enzymes via machine learning

Engineering and characterizing proteins can be time-consuming and cumbersome, motivating previous efforts to develop CRISPR-Cas enzymes with generalist properties to facilitate diverse genome editing applications. However, such universal enzymes have caveats including an increased risk of off-target editing. To enable scalable reprogramming of Cas9 enzymes that are more uniquely suited to specific genomic targets and that have minimized off-target edits, we combined high-throughput protein engineering with machine learning (ML) to derive bespoke proteins with advantageous properties. Via structure/function-informed saturation mutagenesis and bacterial selections, we profiled the protospacer-adjacent motif (PAM) requirements of nearly 1,000 enzymes to train a neural network that relates amino acid sequence to PAM specificity. By utilizing the resulting PAM ML algorithm (PAMmla) to predict the PAMs of 64 million SpCas9 enzymes, we identified efficacious and specific enzymes that outperform evolution-based and engineered SpCas9 enzymes as nucleases and base editors in human cells while reducing genome-wide off-targets. An in silico directed evolution method enables user-directed Cas9 enzyme design, including for custom editors capable of allele-selective targeting of the RHO P23H allele in human cells and mice. Together, PAMmla establishes the feasibility of integrating ML and protein engineering to curate a catalog of SpCas9 enzymes, providing a framework to predict efficient and safe bespoke SpCas9 enzymes that motivate a move away from generalist enzymes for various editing applications.