Generating, modeling, and evaluating a large-scale set of CRISPR/Cas9 off-target sites with bulges

Abstract

CRISPR/Cas9 system is widely used in a broad range of gene-editing applications. While the CRISPR editing technique is quite accurate in the target region, there may be many unplanned offtarget sites (OTS). Consequently, a plethora of highthroughput experimental assays have been developed to measure OTS in a genome-wide manner. Based on these experimental data, computational methods have been developed to predict OTS given a guide RNA and a reference genome. However, these methods are highly inaccurate when considering OTS with bulges due to limited data compared to OTS without bulges. Recently, CHANGE-seq, a newin vitroexperimental technique to detect OTS, was used to produce a dataset of unprecedented scale and quality (more than 200,000 OTS over 110 guide RNAs). In addition, the same study includedin cellulaGUIDE-seq experiments for 58 of the guide RNAs. But, while the CHANGE-seq data included more than 20,000 OTS with bulges, the GUIDE-seq data did not include any OTS with bulges. Here, we fill this gap by generating the most comprehensive GUIDE-seq dataset with bulges, and training and evaluating state-of-the-art machine-learning models that consider OTS with bulges. We first reprocessed the publicly available experimental raw data of the CHANGE-seq study to generate 20 new GUIDE-seq datasets, and more than 450 OTS with bulges among the original and new GUIDE-seq experiments. We then trained various machinelearning models, evaluated their performance on multiple datasets, and demonstrated their state-of-the-art performance bothin vitroandin cellula. Last, we visualized the key features learned by our models on OTS with bulges. Our data and models will be instrumental to any future off-target study considering bulges.