CRISPR Guide RNA Exponential Directed Evolution Synergy with Machine Learning to Improve Gene Editing Efficiency

CRISPR Guide RNA Exponential Directed Evolution Synergy with Machine Learning to Improve Gene Editing Efficiency Abstract: Engineered bacterial Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) systems consist of a single guide RNA (gRNA) and the Cas9 protein forming a ribonucleoprotein complex (RNP) that can scan and cleave DNA. The gRNA consists of a 20 base targeting region complementary to the desired DNA target site, followed by an 80 base scaffold region that binds and activates the Cas9 protein. Unfortunately, high editing efficiency only occurs on a small subset of all possible DNA targets. Editing efficiency is hindered by strong base pairing of the targeting to the scaffold region of the gRNA, and multiple groups have shown that restoring said efficiency can be achieved via low throughput testing of mutations as well as chemical modifications at predicted gRNA misfolding sites. Unfortunately, the vast majority of low efficiency gRNA targets remain unresolved, and while targeting region predictive tools to avoid low efficiency targets exist, no scaffold optimization tools that can fix said efficiency have been developed. Our lab has pioneered a novel high throughput gRNA directed evolution method that can select highly functional scaffolds from libraries of ~1E14 gRNA variants for the Staphylococcus Aureus (sa) CRISPR system, the ortholog with the leading editing efficiency used in AAV gene therapy applications. With the advantage of being encoded for delivery in a single AAV vector. My preliminary data shows improvement of gene editing efficiency via both RNP delivery and plasmid delivery, evolving hundreds of saCas9 gRNA scaffold variants with 10 to 30% variation from the wildtype gRNA scaffold. We hypothesize performing selection with clinically relevant targets will provide the best efficiency scaffolds for these targets, while also expanding our gRNA toolkit. Furthermore, using these scaffolds in a high throughput lentiviral screen, to feed into various structure and sequence-based machine learning and deep learning algorithms, we expect development of a publicly accessible efficiency prediction algorithm leading to improvement of editing efficiency on targets across the human genome. This model will aid in finding the best possible match for improved gRNA efficiency. This will free up time and resources associated with testing gRNAs while also improving the wildtype efficiency paradigm that has ruled the CRISPR gene editing field since its inception. The development of gRNA directed evolution in this Cas9 model, will serve as a proof of concept for the accelerated gRNA efficiency optimization for the use in new CRISPR systems. Furthermore, developing of Strong computational skills, especially in machine learning and deep learning, will enhance an already productive career in academia, where I aim to rapidly optimize CRISPR systems such as Cas9 guided transposons in my postdoctoral studies.