Machine Learning Algorithms in Gene Editing

Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR), as an adaptive immunity mechanism in prokaryotes, is a programmable system that can be used for editing genes of various species. Possibly it can be used to edit genes causing some rare diseases of humans, such as Ichthyosis. CRISPR is associated with the Cas9 protein that cuts a DNA sequence. Cas9 binds to a gRNA that guides Cas9 to an editable site. Although versatile and easy to design, a misguiding gRNA could result wrong edition of genes. Hence, the CRISPR/Cas9 process needs to find ideal gRNAs that can guide Cas9 to on-targets, and avoid off-targets. Various machine learning algorithms play important roles in finding effective gRNAs for the CRISPR/Cas9 by predicting the cleavage efficiency of gRNA. Here, we aim to provide an overview and comprehensive analysis of various machine and deep learning algorithms that are effective in predicting CRISPR gRNA sequences’ on-target activities. Comparison results show that the hybrid approaches combining deep learning and other machine learning algorithms present excellent outcomes. We applied the models to predict the cleavage efficiency of gRNAs from the FLG gene causing Ichthyosis, and the results suggest to use multiple models to obtain consistent and trustworthy results. Although our work does not provide final solutions, it will open opportunities for computational tools to diagnose and treat rare diseases. The datasets and codes are available at https://github.com/ksanyour/MLforGeneEditing.