Discrimination of ranges of closely related RNA targets using CRISPR based detection assay developed using machine learning based optimization

We have developed a CRISPR/Cas based assay able to distinguish between two ranges of closely related RNA targets using two detection channels. This required a pipeline to design RNA guide sets with the right degree of specificity. We tested our approach using SARS-CoV-2 and zoonotic near-neighbor sarbecoviruses. Using pre-existing guide design rules, we utilized a machine learning based model to design and optimize guide sets for specific detection of SARS-CoV-2 and separately to its nearest neighbors. The in vitro testing of the guide sequences has shown that Cas13 assays can tolerate more mismatches than assumed based on previous guide design rules. Mismatches located closer to the 3’ end of the guide and mismatches evenly distributed throughout the guide resulted in a smaller impact on the guide’s ability to activate the Cas enzyme. Modified SHERLOCK assay for detection and discrimination of SARS-CoV-2 and its zoonotic coronaviruses was developed using optimized sets of guides. The final assay was able to classify the targets into three classes 1) SARS-Co-V2, 2) closest known SARS-Co-V2 near-neighbor BANAL-236 and 3) the remaining zoonotic near-neighbors. This approach provides value through early detection of natural and engineered variants.