Benchmarking and integrating genome-wide CRISPR off-target detection and prediction

Abstract

Systematic evaluation of genome-wide Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) off-target profiles is a fundamental step for the successful application of the CRISPR system to clinical therapies. Many experimental techniques and in silico tools have been proposed for detecting and predicting genome-wide CRISPR off-target profiles. These techniques and tools, however, have not been systematically benchmarked. A comprehensive benchmark study and an integrated strategy that takes advantage of the currently available tools to improve predictions of genome-wide CRISPR off-target profiles are needed. We focused on the specificity of the traditional CRISPR SpCas9 system for gene knockout. First, we benchmarked 10 available genome-wide off-target cleavage site (OTS) detection techniques with the published OTS detection datasets. Second, taking the datasets generated from OTS detection techniques as the benchmark datasets, we benchmarked 17 available in silico genome-wide OTS prediction tools to evaluate their genome-wide CRISPR off-target prediction performances. Finally, we present the first one-stop integrated Genome-Wide Off-target cleavage Search platform (iGWOS) that was specifically designed for the optimal genome-wide OTS prediction by integrating the available OTS prediction algorithms with an AdaBoost ensemble framework.

Molecular and Computational Strategies to Increase the Efficiency of CRISPR-Based Techniques

“…The rapid progress of experimental procedures implementing CRISPR/Cas technology in plants over the past decade has been accompanied by equally impressive advances in the computational methods for gRNA design and off target prediction ( Wang et al, 2017b , 2020b ; Lowder et al, 2018 ; Hajiahmadi et al, 2019 ). As genetic transformation methods and CRISPR implementation grew, the emerging algorithm developments for accurate target prediction revealed increasingly complex facets of the underlying biology, from sequencing composition to epigenetic regulation ( Wang et al, 2019 ; Yan et al, 2020 ). At the same time, rapid growth has forced constant reevaluation of the underlying algorithms and statistical models used by these computational tools when implemented in CRISPR studies ( Sledzinski et al, 2020 ).…”

“…Over the last couple of years, solid evidence has shown that hypothesis-driven and learning-based strategies outperform alignment-based strategies ( Wang et al, 2020c ; Yan et al, 2020 ; Zhang et al, 2020 ). However, the popularity gained by artificial intelligence has positioned learning-based methods as the leading strategies to increase efficiency efforts.…”

“…Several machine and deep learning-based methods have been developed to predict CRISPR on-target activity. However, most of them use data sets exclusively for animal models ( Chuai et al, 2018 ; Kim et al, 2019 ; Yan et al, 2020 ; Xiang et al, 2021 ). Even though animals and plants share genetic and epigenetic features, models specifically for plants are needed.…”

“…Even though animals and plants share genetic and epigenetic features, models specifically for plants are needed. Based on their statistical methods ( Feng et al, 2017 ; Lin and Wong, 2018 ; Chen et al, 2019b ; Wang et al, 2019 , 2020c ; Liu et al, 2020 ; Padilha et al, 2020 ; Yan et al, 2020 ; Zhang et al, 2020 ; O’Brien et al, 2021 ), most of the models developed can be roughly classified into six categories: (i) CNN: Convolution Neural Network; (ii) L1-Reg: L1-Regression; (iii) SVM: Support Vector Machine; (iv) RF: Random Forest; (v) GBRT: Gradient Boost Regression Tree; and (vi) SVM (C): using SVM to classify (+1 represents high activity, −1 represents low-activity).…”

A systematic evaluation of data processing and problem formulation of CRISPR off-target site prediction

“…less than 10 sgRNAs per dataset with only tens of active off-target sites in total [33], resulting in inaccurate models for off-target prediction. For example, in a recent study, a state-of-the-art ensemble model was trained on a benchmark dataset and tested in cross-validation [34]. We gauged the model's performance, and it achieved an average area under the precision-recall curve (AUPR) over 16 sgRNA targets of only 0.31 ± 0.2.…”