A predictable and precise method to edit pathogenic genetic variants using a machine learning approach is reported in a paper published online this week in Nature. This work suggests new capabilities for the study and potential treatment of genetic diseases.
CRISPR-Cas9 has revolutionized genome editing for research, but ensuring the accuracy of the technique is of vital importance. So-called DNA ‘templates’ are usually used in CRISPR-Cas9 genome editing to enable accurate DNA repair or introduce a specific DNA sequence into the genome, and DNA repair without these templates is thought to be less accurate.
Now, Richard Sherwood and colleagues have demonstrated precise template-free Cas9 editing by developing a method to predict genome repair outcomes using machine learning. The authors used a library of nearly 2,000 paired Cas9 guide RNAs (gRNAs) and human DNA target sites to train the inDelphi machine learning model, which identified that 5-11% of Cas9 gRNAs targeted to the human genome can induce a single, predictable repair outcome in over 50% of instances (so-called ‘precise-50’). inDelphi could also identify and predict suitable pathogenic genetic variations to target using template-free Cas9 editing, including some that it had been thought could not be targeted in this way.
Finally, the authors experimentally confirmed in human cells that nearly 200 pathogenic variants related to three diseases - Hermansky-Pudlak syndrome, Menkes disease and familial hypercholesterolemia - could be accurately edited and repaired to precise-50 standards.
These findings establish an approach for precise, template-free genome editing.