Using deep learning to find better CRISPR-Cas13d tools for targeting RNA

Over 127,000 RfxCas13d (CasRx) guide RNAs were analyzed to develop an efficient RNA targeting prediction algorithm.

• Seven machine learning models were evaluated to create a guide efficiency prediction algorithm validated across various human cell types.
• Preferred sequence patterns and secondary features for effective guide RNAs were identified through deep learning model interpretation.
• Screening of 46 novel Cas13d orthologs revealed that DjCas13d has low cellular toxicity and high specificity.
• DjCas13d maintained high specificity even when targeting abundant transcripts in sensitive cell types such as stem cells and neurons.
• The guide efficiency model developed for RfxCas13d was successfully applied to DjCas13d, demonstrating the potential of combining machine learning with ortholog discovery.

Simplified